This article is licensed under CC-BY 4.

[Link]/jcim Perspective

Targeted Protein Degradation: Advances, Challenges, and Prospects

for Computational Methods
Barmak Mostofian, Holli-Joi Martin, Asghar Razavi, Shivam Patel, Bryce Allen, Woody Sherman,
and Jesus A Izaguirre*

Cite This: J. Chem. Inf. Model. 2023, 63, 5408−5432 Read Online

ACCESS Metrics & More Article Recommendations

ABSTRACT: The therapeutic approach of targeted protein

degradation (TPD) is gaining momentum due to its potentially
superior effects compared with protein inhibition. Recent advance-
ments in the biotech and pharmaceutical sectors have led to the
development of compounds that are currently in human trials, with
some showing promising clinical results. However, the use of
computational tools in TPD is still limited, as it has distinct
characteristics compared with traditional computational drug
design methods. TPD involves creating a ternary structure
(protein−degrader−ligase) responsible for the biological function, such as ubiquitination and subsequent proteasomal degradation,
which depends on the spatial orientation of the protein of interest (POI) relative to E2-loaded ubiquitin. Modeling this structure
necessitates a unique blend of tools initially developed for small molecules (e.g., docking) and biologics (e.g., protein−protein
interaction modeling). Additionally, degrader molecules, particularly heterobifunctional degraders, are generally larger than
conventional small molecule drugs, leading to challenges in determining drug-like properties like solubility and permeability.
Furthermore, the catalytic nature of TPD makes occupancy-based modeling insufficient. TPD consists of multiple interconnected yet
distinct steps, such as POI binding, E3 ligase binding, ternary structure interactions, ubiquitination, and degradation, along with
traditional small molecule properties. A comprehensive set of tools is needed to address the dynamic nature of the induced proximity
ternary complex and its implications for ubiquitination. In this Perspective, we discuss the current state of computational tools for
TPD. We start by describing the series of steps involved in the degradation process and the experimental methods used to
characterize them. Then, we delve into a detailed analysis of the computational tools employed in TPD. We also present an
integrative approach that has proven successful for degrader design and its impact on project decisions. Finally, we examine the
future prospects of computational methods in TPD and the areas with the greatest potential for impact.

1. INTRODUCTION While traditional small molecule drug discovery has primarily

1.1. Historical Overview of Degradation. Protein focused on directly controlling protein activity, harnessing the
degradation plays a crucial role in cellular regulation.1 The endogenous protein destruction machinery within cells to
26S proteasome is the primary enzyme in the ubiquitin- selectively degrade key drivers of human diseases holds great
dependent protein degradation pathway, which is the main potential for drug discovery.
focus of this Perspective. The proteasome, a nanomachine that The discovery of cells’ ability to degrade proteins through a
relies on energy, breaks down proteins that are covalently nonlysosomal pathway dates back to the early 1980s.7
“tagged” with ubiquitin by an E2 ligase. This ligase usually Although the mechanism was initially unclear, researchers
participates in a larger macromolecular assembly responsible soon identified a small regulatory protein called ubiquitin,
for identifying proteins to be targeted for degradation through which was covalently attached to the protein before
a degron recognition motif. The ubiquitin−proteasome degradation. Subsequently, it was found that three enzymes
pathway is vital for controlling protein levels in cells and, (E1, E2, and E3) played a role in the ubiquitination process.8
therefore, essential for life.2 Since the early 2000s, several
researchers have endeavored to exploit this pathway using Received: April 27, 2023
molecules that induce proximity and can be leveraged for Published: August 21, 2023
therapeutic purposes to degrade a protein of interest (POI).3−6
A degradation strategy offers multiple advantages over small
molecule inhibitors, including targeting nonfunctional sites on
the POI and the catalytic mechanism by which it operates.

© 2023 American Chemical Society [Link]

5408 J. Chem. Inf. Model. 2023, 63, 5408−5432
Journal of Chemical Information and Modeling [Link]/jcim Perspective

Figure 1. Example of a heterobifunctional degrader molecule. The structure of ACBI1 is depicted, a prominent degrader of SMARCA2 that recruits
von Hippel−Lindau (VHL) E3 ligase. The warhead (red), linker (green), and E3 ligand motifs (blue) are highlighted, illustrating the typical
composition of a degrader molecule.

The large protein assembly (20S proteasome, part of the 26S meaning that static representations might be inadequate for
proteasome) responsible for protein degradation was charac- correlating structure to function.25 The formation of the
terized9 and crystallized10 in the 1990s. These groundbreaking ternary complex enables the ubiquitination of the POI, which
discoveries inspired drug researchers to investigate the generally involves ubiquitin-conjugating enzymes (E2s).26
possibility of inducing the proximity between a protein of However, not all ternary complexes lead to the same level of
interest and an E3 ligase, leading to growing efforts in POI ubiquitination due to the spatial and kinetic aspects of the
academia and industry to develop molecules capable of ubiquitination process. Lastly, once ubiquitinated, the POI is
harnessing degradation biology for therapeutic purposes. The prepared for degradation by the proteasome.27
first heterobifunctional degrader molecule was reported in Solving crystal structures of ternary complexes has been
200111 and in 2019, the first designed degrader molecule achieved,23 but converting these structures into actionable
entered human clinical trials.12 This progress has sparked a designs is not a simple task. Empirical rules created for
surge of interest and investment in the field of targeted protein traditional small molecules may not be readily applicable as
degradation (TPD). Initial research efforts in TPD focused on most heterobifunctional degraders have a molecular weight
well-validated cancer targets such as estrogen receptor (ER)13 significantly above 500 Da. Computational chemistry tools,
and androgen receptor (AR).14 However, the TPD strategy such as small molecule docking, are not designed to
extends beyond specific indications or therapeutic areas.15 As a simultaneously bind a degrader molecule and two proteins.
result, TPD research now covers a wide range of targets and Additionally, the relationship between the ternary structure
therapeutic areas, including notoriously challenging targets like and degradation efficiency remains unclear. Some researchers
KRAS16 and MYC,17 as well as lesser-known targets such as argue that more rigid linkers are better,28 while others suggest
IRAK4,18 STAT3,19 and Tau.20 that flexibility is superior.29 The answer is likely dependent on
The process of targeted protein degradation by a degrader the specific system’s nuances, with numerous factors
molecule consists of several steps that can impact the contributing to the design of an effective degrader molecule.
degradation efficiency. Some of these steps resemble those of 1.2. Heterobifunctional Molecules as Therapeutics.
traditional small molecule therapeutics, while others are Therapeutic molecules for targeted protein degradation can
unique. For instance, like traditional small molecules, a generally be classified into “glues” and “heterobifunctionals”,
degrader must possess an adequate solubility and permeability although in reality there is a spectrum between these two
to reach the protein of interest (POI). However, the extremes. Glues resemble traditional small molecules in shape
mechanistic steps diverge once the degrader reaches the and size that, upon binding, function as “adhesives” and
POI. First, TPD is a catalytic process in which a single convert the target protein into a so-called “neo-substrate” for
molecule can cause the degradation of numerous POI the E3 ligase. Conversely, heterobifunctional degraders, as
copies.21,22 The process commences when the degrader visualized in Figure 1, are larger and consist of two protein-
induces proximity between a POI and an E3 ligase by forming binding components connected by a linker motif, usually
ternary complexes.23 These complexes can involve non-native inducing a more flexible ternary complex. In this Perspective,
interactions between the two proteins as the therapeutic we focus primarily on heterobifunctional degrader molecules.
strategy often involves repurposing the E3 ligase to act on non- In terms of nomenclature, we will refer to the molecular
native substrates. The strength and nature of these protein− motif binding the protein of interest as the “warhead”, the
protein interactions play a crucial role in degradation motif binding the E3 ligase as the “E3 ligand”, and the “linker”
efficiency.24 Moreover, the ternary complex is typically flexible, connecting the two motifs. This topology presents both
5409 [Link]
J. Chem. Inf. Model. 2023, 63, 5408−5432
Journal of Chemical Information and Modeling [Link]/jcim Perspective

opportunities and challenges in the therapeutic design. On one activity relationship (QSAR) models, and even if such data
hand, it is possible to rapidly design degrader molecules by existed, the model-building methods might need to change.
connecting a known warhead to an E3 ligand using a flexible For example, fingerprint representations may be insufficient for
linker, as demonstrated in numerous examples.30−33 However, constructing meaningful predictive models, and the quantity of
the linker’s nature can significantly affect degradation, likely data required to build high-quality models might be
due to a combination of factors that will be discussed in greater significantly larger for heterobifunctionals due to their size
detail below. As a result, while hit finding can be relatively and conformational flexibility. Furthermore, traditional phys-
straightforward in TPD programs when there are known POI iochemical properties might not directly inform project
warheads, designing a degrader remains a challenging task. decisions, given the complex, multistep catalytic process.
First, enhancing properties such as solubility and perme- While the scarcity of data will improve over time, it is difficult
ability is often more challenging for heterobifunctional to predict when enough information will be available to make
degraders than for traditional small molecules. This is due to reliable predictions on various properties for heterobifunc-
the lack of substantial data and the innate complexity of these tionals.
larger, more flexible molecules. Moreover, connecting chemical 1.4. Assays for TPD. TPD shares some similarities with
structures to degradation is difficult, as binding is necessary but traditional drug discovery such as the necessity of binding, but
not sufficient to induce ubiquitination, and enzymatic it also exhibits numerous differences. TPD calls for a more
deubiquitination may interfere. Additionally, degraders could diverse approach (both computationally and in the wet lab)
bind to each protein-binding partner individually leading to the utilizing various tools and unconventional ways of conceptu-
concentration-dependent hook effect, i.e., the formation of alizing the problem. The lack of a direct correlation between
binary complexes that compete with the productive ternary binding and degradation necessitates that cellular assays be
complexes and thus negatively impact the degrader potency conducted earlier in the process and serve as a primary assay.
and also complicate toxicological assessments.34−36 Further- Additionally, permeability can sometimes fall below the
more, there are only a few computational tools specifically detectable limit by using traditional small molecule methods,
designed for predicting and designing the properties of further emphasizing the importance of early cellular assays.
heterobifunctional molecules. To predict the activity of these Since the primary objective of TPD is to reduce protein
molecules, it is beneficial, and sometimes essential, to consider levels, it is crucial to monitor these levels within the cell,
factors beyond binding, such as the conformational landscape typically through proteomic techniques. This monitoring
of the ternary complex ensemble, which is crucial for becomes central to the project rather than just being part of
understanding functionality. Lastly, while several molecules exploratory biology. In fact, due to its catalytic nature, TPD
based on heterobifunctional design are currently in clinical demands a new way of thinking about target product profiles
trials, there are no approved heterobifunctional degraders. As a (TPPs), as the event-driven degradation mechanism can still
result, there is limited information about potential toxicities exhibit high efficacy at lower doses compared to the
and other liabilities associated with this emerging class of occupancy-driven pharmacology of traditional inhibitors.
molecules. The intricacies of TPD suggest that successful research
1.3. Traditional Computational Approaches are not groups in this area will benefit from a more diverse workforce
Suited to Heterobifunctional Molecules. Targeted protein and the proper integration of various skill sets. Further details
degradation necessitates a novel approach to small molecule on TPD assays can be found in Section 2.
drug discovery, shifting from static structures that function 1.5. Overview of This Perspective. Although TPD
primarily via occupancy-driven pharmacology to conforma- research has thus far been successful without substantial
tional stabilization within a complex free energy landscape to computational investments, the development and deployment
achieve catalytic turnover. X-ray structures are often of appropriate tools can have a significant impact. Emerging
inadequate for understanding structure−activity relationships examples show that traditional computational tools are being
(SAR), given the dynamic nature of the ternary complex and adapted for TPD applications, such as ternary complex
its implications for downstream ubiquitination. As previously docking, which will be covered in this Perspective. However,
mentioned, the warhead−linker−ligand configuration of in our opinion, the most significant impact will stem from new
heterobifunctional molecules, along with the importance of techniques specifically designed for TPD applications, such as
the ternary structure, poses challenges for traditional computa- molecular simulations of various TPD steps, which we will
tional methods. discuss in detail.
For instance, conventional small molecule docking tools can To develop new tools, it is crucial to understand the physical
be used to position the warhead and E3 ligand to the POI and steps associated with TPD. We will start by examining the
E3 ligases, respectively, but they are insufficient for predicting targeted protein degradation process and the range of
ternary structures. Likewise, traditional protein−protein experimental assays used to explore it. Subsequently, we will
docking tools do not consider small molecule sampling in discuss how each physical step can be modeled using different
conjunction with global (translational and orientational) computational approaches, with a focus on predicting proper-
protein sampling. In essence, predicting the ternary structure ties like solubility and permeability, multiscale simulations of
combines some of the most challenging aspects of both small various steps, and mathematical modeling of degradation
molecule docking and protein−protein docking into one events. As the ultimate goal of TPD research is to design
problem, although it is possible to simplify the search problem innovative therapeutics, we will explore what an integrative
in certain cases by applying constraints. design process for heterobifunctional degraders might entail,
Predicting other drug-like properties, such as solubility, showcasing some recent results that demonstrate the impact of
permeability, bioavailability, and clearance, also presents computational methods in degrader design and the value of
challenges. Currently, there are minimal data available to integration with experiments. Lastly, we discuss potential areas
create machine learning (ML) or quantitative structure− for future investment and impact.
5410 [Link]
J. Chem. Inf. Model. 2023, 63, 5408−5432
Journal of Chemical Information and Modeling [Link]/jcim Perspective

Figure 2. Physical steps of the TPD process. A degrader must be soluble in aqueous environments for its systemic circulation and distribution
(“Solubility”), but it should also be lipophilic enough for passive transport into the cell (“Permeability”). Inside the cell, the degrader binds through
its warhead to the target protein of interest and through its E3 ligand motif to an E3 ligase (“POI or E3 Ligase Binding”), yielding a so-called
ternary complex (“Ternary Structure Formation”) that induces the “Ubiquitination” of the POI in a supramolecular assembly (e.g., the Cullin−
RING ligase). “Degradation” is then achieved by the cell-innate ubiquitin−proteasome pathway.

Table 1. Summary of Assays and Techniques That Have Been Employed for Targeted Protein Degradation
Category Assaya Readout References
28,37−39
Target engagement assays Surface plasmon resonance (SPR)v Binding affinity and kinetics through changes in light reflection
28,40−43
Isothermal titration calorimetry (ITC)v Binding affinity, stoichiometry, thermodynamic quantities from heat
generation/consumption
28,37,41,43−45
Fluorescence polarization (FP)v,c Binding affinity and selectivity by changing fluorescence signal
28,39,45
Ternary complex Time-resolved Förster resonance energy transfer Formation of ternary complex by changes in fluorescence intensity
formation assays (TR-FRET)v,c
28,38,40,41
Amplified luminescent proximity homogeneous Ternary complex detection through luminescence signal
assay (ALPHA)v,c
Proximity ligation assay (PLA)c Detection of ternary complexes by fluorescence or luminescence �
28,37−47
Functional cellular assays Western blotv Quantification of target protein levels by specific antibodies
28,41,48,49
CRISPR/Cas9-mediated gene editingc TPD effects compared to genetic loss of the target protein
50−52
Cell-based phenotypic assaysc Biological consequences of degrading the target protein
a
The superscripts indicate if the corresponding experiment is usually performed in vitro (v) or as a cell-based assay (c).

2. OVERVIEW OF THE TPD PROCESS�FROM operations does not matter in terms of binding to the POI
ADMINISTRATION TO PROTEIN DEGRADATION or the E3 ligase first, although kinetically, there may be
The TPD process involves a series of steps, all of which must differences.
occur for successful degradation. We describe the process from A variety of different experimental techniques and assays
when a drug enters the body to when the targeted protein is have been developed to characterize each of the TPD steps. As
degraded, with relevant assays interspersed. summarized in Table 1, some of the most commonly used
Figure 2 shows a graphical depiction of the primary steps assays can be broadly categorized into (1) target engagement
involved in the TPD process, from administration of a assays, which are designed to measure the ability of a degrader
heterobifunctional degrader to theproteasomal degradation to bind to the POI or E3 ligase, (2) ternary complex formation
event. For a molecule to be efficacious as a drug it must reach assays, which focus on evaluating the formation of the ternary
the target of interest. For a degrader molecule this still holds complex, and (3) functional cellular assays, which evaluate the
true, although the amount required is likely to be significantly ability of a degrader to induce degradation of the target protein
lower. The molecule must be soluble to get distributed in the and its downstream effects on cellular function. A combination
blood and permeable enough for at least one copy of the of these assays is often employed to characterize degraders,
degrader to get into the cell. Once in the cell, the molecule assess their specificity, and determine their potential
must reach the protein of interest, bind to the two partners therapeutic utility in discovery projects. In the following
(POI and E3), and induce ubiquitination. Compared with paragraphs, we discuss these and several other promising
traditional small molecule drugs, degraders need a lower techniques and the caveats associated with TPD projects.
concentration in the cell because they operate by an event- As with the development of any small-molecule drug,
driven mechanism.21,22 Thermodynamically, the order of monitoring biopharmaceutical properties, e.g., bioavailability
5411 [Link]
J. Chem. Inf. Model. 2023, 63, 5408−5432
Journal of Chemical Information and Modeling [Link]/jcim Perspective

and pharmacokinetics, during the discovery of degraders is physicochemical properties in polar and nonpolar solutions,
crucial to guide the decision-making process and to ensure the thus providing a structural basis for their permeability
development of active and potent degraders. Several studies differential. The recent development of new technologies and
have offered insight into degrader metabolism and pharmaco- the creative use of existing tools for the analysis of both the
kinetics,53−55 concluding, based on solubility, biotransforma- solubility and permeability of degraders will advance our
tion, and elimination data as well as on nonenzymatic stability knowledge of distinct features that govern their bioavailability.
analysis, that heterobifunctional degraders may well achieve As mentioned earlier, once degraders reach their targets,
acceptable drug metabolism and pharmacokinetics (DMPK) their mode of action is different from that of inhibitors. In
properties, although optimization strategies are required. particular, degraders induce two binding events. The
Favorable aqueous solubility and membrane permeability are corresponding binding kinetics are usually assessed by
key attributes in the development of therapeutics56 and their surface−plasmon resonance (SPR) that informs on the lifetime
mutual interplay, with reference to drug delivery and or stability of the degrader-induced ternary complex and the
distribution, is well known.57,58 The correct estimation of associated (un)binding rate constants.37 Since the protein is
these properties is critically important for ADMET (Absorp- typically attached to a surface through a tag, the native binding
tion, Distribution, Metabolism, Elimination, Toxicity) screen- properties of the target protein may be disrupted in SPR.
ing and lead optimization, which is discussed in greater detail Nevertheless, these measurements are instrumental in
in Section 3.1 in the context of predicting and modeling characterizing a degrader’s binding affinity to both the POI
different aspects of degraders. Importantly, the assessment of and the ligase and hence its degree of binding cooperativity,
oral degrader absorption is more complicated than that of i.e., the increase in a degrader’s affinity to either binding
small-molecule drugs because, similar to macrocyclic or partner in a ternary complex compared to the corresponding
peptidomimetic compounds, degraders exhibit a certain affinity in a binary (nonternary) structure. Cooperativity is a
“chameleonic behavior” in response to different environments, key parameter to determine degrader specificity.38
enabling them to passively permeate cell membranes and, yet, Competitive-binding and proximity-based assays can also
be sufficiently dissolved under aqueous conditions.59,60 Due to probe the degree of degrader−POI engagement or ternary
their relatively large structure, heterobifunctional degraders are complex formation (Table 1), thus allowing the determination
assigned to the so-called “beyond rule of five” (bRo5) chemical of binding affinities and cooperativity. These techniques
space,61,62 i.e., the accurate estimation of their properties is measure changes in the polarization of a fluorescently labeled
strongly hampered. Therefore, recent lessons learned on the target protein upon binding to a degrader, as in fluorescence
molecular features of bRo5 compounds will prove highly polarization (FP),85 or detect intensity changes, upon ternary
valuable for degrader design.63−66 complex formation, in the fluorescence or luminescence signal
A variety of experimental physicochemical profiling emitted from beads tagged to the POI or E3 ligase, as in time-
methods67,68 are usually applied to characterize properties resolved Förster resonance energy-transfer (TR-FRET)86 or in
such as polarity, lipophilicity, ionization, or stability, which are the amplified luminescent proximity homogeneous assay
major determinants of a compound’s bioavailability.69 For (ALPHA).87 In the proximity ligation assay (PLA), a
instance, partition assays or reverse-phase HPLC70,71 measure technology that is gaining popularity in TPD research, the
the lipophilicity, usually reported as the octanol−water signal is even further amplified through antibody-attached
partition coefficient, Log P, i.e., the equilibrium distribution oligonucleotide hybridization.88 These methods have been
of a compound between water and octanol, or, as the used in TPD studies for the identification of selective
distribution coefficient, Log D, for ionizable species, degraders and for the high-throughput screening of degrader
respectively. designs.39−41,89 In particular, ALPHA has a high dynamic range
Promisingly, systematic studies on the aqueous solubility of and signal-to-noise ratio and, when applied in a titration
degraders are currently emerging in the literature.72,73 To experiment (usually set up as an immunoassay called
directly measure the thermodynamic aqueous solubility of a AlphaLISA), it yields relative populations of ternary complexes
substance, Log S, traditional saturation methods (in combina- as a function of degrader concentration and is routinely
tion with LC/MS analysis) or potentiometric methods are employed as a tool in degrader design projects (see Section
often applied.74 In particular, for degraders that contain 4.2).
ionizable sites, the pH-dependent solubility can be assessed Complementary to the methods described above, isothermal
and the intrinsic solubility, Log S0, i.e., that of the neutral titration calorimetry (ITC) is a label-free technique and thus
species, reported.75 Also, for poorly soluble compounds like does not require tagging or immobilization of the analytes. ITC
many degraders, dissolution profiles from amorphous solid or measures the heat released or absorbed during the binding of a
liquisolid formulations are sometimes derived.76 degrader to its target protein, providing information on
Since different degrader chemotypes can penetrate different binding affinity, stoichiometry, and the changes in enthalpy
cell types, it has long been argued that their cellular uptake and entropy, although at lower throughput and higher protein
occurs through passive diffusion. Recently, common label-free demands. The ITC-guided optimization of degraders has been
methods that measure the cell permeability, such as the parallel reported for different POI−ligase pairs, e.g., Tau-KEAP142 and
artificial membrane permeability assay (PAMPA)77,78 or the BRD4-VHL.40
Caco-2 cell monolayer assay,79 have shed light on degrader Alternatively, NMR spectroscopy, which is based on
permeabilities.80,81 To increase the sensitivity, a HaloTag- chemical shift perturbations upon ligand binding, is fairly
based assay, called the chloroalkane penetration assay (CAPA), sensitive to binding signals90 and has been used for fragment
or even system-specific intracellular reporter-based assays have screening on protein surfaces,91,92 in particular to identify
been designed that helped rank degraders based on their binding pockets on the von Hippel−Lindau (VHL) E3 ligase93
permeability.82,83 In an interesting approach, Atilaw et al.84 and to aid in the optimization of its inhibitors.94 Recently,
applied NMR spectroscopy to study degraders with different Castro and Ciulli reported an NMR assay that facilitates the
5412 [Link]
J. Chem. Inf. Model. 2023, 63, 5408−5432
Journal of Chemical Information and Modeling [Link]/jcim Perspective

Figure 3. Properties of interest and computational approaches in TPD. The multistep process of TPD entails a diverse set of modeling techniques
developed for the prediction of degrader physicochemical properties, degrader-induced ternary structures, and the degree of target ubiquitination
and degradation. Higher-level properties such as pharmacokinetics (PK) and pharmacodynamics (PD) could encompass all of these properties,
plus additional properties (e.g., tissue localization of the degrader, metabolites, and off-target effects). This figure is intended to be suggestive and
not exhaustive. The approaches described here are software-agnostic and could be performed with a variety of open-source and commercial
software packages.

evaluation of cooperativity in degrader-induced ternary binding, or consequently, ternary complex formation. These
complexes.95 two methods characterize the protein binding interface in a
X-ray crystallography has been used to resolve the structures fairly complementary fashion,104 as different regions are
of entire ternary complexes,96 or those of bound warhead− labeled (side chains versus backbone) and, as a result, different
POI44 or E3 ligand−ligase,97 revealing distinct interactions in time scales can be assessed (ns−μs versus ms−s).105,106 In
atomic detail and providing invaluable structural knowledge for particular, HDX-MS has already been applied to elucidate
rational degrader design. However, X-ray crystallography is a protein sites that contribute to ternary complex formation, and
very challenging technique and hence cannot always be the information provided can easily augment follow-up
applied, particularly not for relatively large and, yet, dynamic modeling and simulation analyses.107,108 In our opinion,
ternary complexes that are difficult to co-crystallize. This issue these structural MS techniques can be ideally supported by
can be remedied by cryogenic electron microscopy (cryo-EM) other chemical footprinting or even site-directed mutagenesis
that solves large macromolecular structures, including ternary experiments in which the role of distinct residues at the
complexes.98 Cryo-EM is generally considered to become an binding interface can be intensely explored.
essential tool in drug discovery,99 and we believe that it will Mass spectrometry techniques are pivotal for quantitative
play an important role in the discovery of novel degraders. proteomics to characterize protein modifications and thus
Other (label-free) tools that have been used to provide assess target ubiquitination and degradation.109,110 For
insight into degrader−protein engagement and ternary instance, when coupled to in vitro ubiquitination assays, MS
complex formation or ubiquitination include size-exclusion analysis has identified distinct degrader-induced ubiquitination
chromatography (SEC), that can compare the degree of sites and (poly-)ubiquitination linkage types,40,111,112 although
complex formation among different degraders,100 native mass the coverage of ubiquitinated lysines may be incomplete in
spectrometry (MS), that has helped explore degrader these experiments. Recent advances in live-cell ubiquitinomics
selectivity and specificity,101 and differential scanning calorim- involve the transfection of HeLa cells with the tagged POI and
etry, that, when applied as a cellular thermal shift assay ubiquitin, followed by immunoprecipitated ubiquitin pull-
(CETSA),102 exploits the differential in the thermal stabiliza- down and electrophoretic or LC/MS analysis.38,108 MS-based
tion upon ternary aggregation in a cellular environment.38 global proteome profiling can quantitate the abundance of
The Western blot immunoassay, after separation of protein target proteins, i.e., their turnover and resynthesis rates, and
samples by gel electrophoresis, is frequently used to measure therefore the effective degrader response28,39,44 and possibly
the target protein levels inside the cell, thus assessing the identify off-target effects.113 Proteomics approaches are
degrader activity. Although capillary electrophoretic immuno- believed to play an outstanding role in TPD research.114
assays are believed to be less error-prone,103 Western blot, To monitor the intracellular degradation, luminescence-
coupled with other assays, has become a standard tool to probe based reporter assays with endogeneous HiBiT-tagged proteins
target protein ubiquitination and degradation.28,37−47 have been developed that yield degradation profiles, from
Both fast photochemical oxidation of proteins (FPOP) and which maximum degradation levels (Dmax) as well as the half-
hydrogen−deuterium exchange (HDX) are MS-based foot- maximum degradation concentration (DC50) can be derived to
printing methods that report on the solvent accessibility of estimate the potency of a degrader molecule,48 which is
proteins, which is altered upon conformational changes, ligand instrumental in degrader design (see Section 4.2). As a matter
5413 [Link]
J. Chem. Inf. Model. 2023, 63, 5408−5432
Journal of Chemical Information and Modeling [Link]/jcim Perspective

of fact, variants of this technology are applied to characterize to current QSAR methods, which often exploit topological
the different steps in TPD, such as target engagement, ternary descriptors, simulation trajectories are information-rich, ideally
complex formation, and target ubiquitination,115 providing containing a full mechanistic and dynamic description of the
complementary kinetic information to the other experimental system, providing useful information on the thermodynamics
approaches discussed. and kinetics of a drug−target binding process140 in addition to
To evaluate degrader activity phenotypically, the target extracting 3D descriptors, such as the radius of gyration or
phenotype is often compared to those from CRISPR/Cas9- intramolecular hydrogen bonding. We elaborate in the
mediated protein knockouts or from RNAi screens.48,49 Recent following paragraphs why we expect MD simulations to be
techniques even employ reporter genes to reflect the integral in many aspects of TPD modeling and degrader
expression level of target proteins. Measuring the intracellular design.
downstream effect of TPD, such as changes in cell viability,51 3.1. The Prediction of Degrader Physicochemical
proliferation,50 or morphology,52 is a functional readout that Properties Is Challenging. Traditionally, the computational
can provide insights into the biological consequences of assessment of a drug’s thermodynamic aqueous solubility
degrading the target protein. ranges from predictions of properties such as melting points
The variety of methods being deployed in TPD projects (Tm)141 or partition coefficients142 and the application of
attests to the complexity of the TPD process and mirrors the QSPR methods143,144 to the estimation of solvation free
fact that multiple molecular events must take place in order to energies145−147 and solubility parameters from molecular
achieve degradation. In combination with molecular modeling, simulations.148−150 In recent years, ML methods for the
this large suite of technologies has already helped advance our prediction of solubilities in water and in organic solvents have
knowledge and understanding of the different TPD steps for become popular.151−153 Recent applications of predictive
distinct POI−ligase pairs. For the sake of completeness, we models to degrader molecules, or similar high-molecular-
would like to note that multiple other approaches, not listed weight compounds in the bRo5 space, suggest that the same
above, have also been recently applied.116 Furthermore, the rules developed for small molecules cannot be readily
current toolbox is rapidly expanding as new innovative transferred.154,155 In particular, the structural flexibility of
methods are being developed to provide greater detail about large heterobifunctional degrader molecules requires model
the TPD-related processes. revisions such that certain parameters that support chameleo-
nicity (the ability of a molecule to adopt both hydrophobic and
3. MODELING THE TPD PROCESS hydrophilic 3D conformations) are combined with those that
Despite the breadth of biochemical and biophysical tools favor oral bioavailability (e.g., the distribution coefficient, Log
developed and adapted for exploring TPD, an integrated D).156,157 Moreover, the dynamic nature of features, such as
approach to model the different steps can lead to more holistic the changing exposure of surface polarity, seems to better
predictions of the entire degradation process.117 Figure 3 capture the solubility (and permeability) of bRo5 compounds
shows a summary of properties that help characterize different than traditional two-dimensional descriptors do.29
stages of the TPD process and the state-of-the-art modeling To systematically distinguish degraders from nondegraders
techniques applied for their study. in their solubility, Jiménez et al.72 trained a decision tree
While pharmacokinetic−pharmacodynamic (PK−PD) and classifier that relies on correlations between a set of
other quantitative models formulate the multistep TPD experimentally obtained lipophilicity descriptors and several
mechanism mathematically, it is possible to model some in silico solubility predictors based on the structure of ∼20
aspects of the process on a molecular level using quantitative heterobifunctional molecules. Furthermore, they also examined
structure−activity/structure−property relationships (QSAR/ the effect each of the three different motifs within a
QSPR)118 as well as molecular modeling and simulation. heterobifunctional molecule, i.e., the warhead, the linker, and
In QSAR/QSPR, multivariate regression and classification the E3 ligand, could have on a degrader’s solubility. Despite
analyses, augmented by a variety of machine learning (ML) the fact that they derive preliminary guidelines to identify
methods,119−122 are applied to predict physicochemical traits, soluble degraders, this study reveals, as the authors admit, the
that determine ADMET-related effects, based on a set of complexity of the task of designing orally bioavailable
molecular descriptors, typically representing the chemical degraders.
structure of a compound at the level of its structural formula Similar to the prediction of solubility, QSPR methods have
(2D) or its conformation (3D).123 As a result of increased also been established between results from experimental
training data availability and advances in deep learning permeability screens, e.g., PAMPA, and structural descriptors
methodologies, graph convolution models that learn a such as the intramolecular hydrogen bonding,158 solvent-
molecule’s structural representation have recently been accessibility,159 or molecular size of the permeant.160 Often,
developed for property prediction.124−126 However, as noted these models use an expression for the water−membrane
earlier, for most “beyond rule of 5” (bRo5) compounds and partitioning160 or free energy of transfer.161 To sample
particularly for heterobifunctional degraders, the accurate sufficient conformations, MD simulations are routinely
prediction of physicochemical properties is quite challenging. applied162,163 and, in particular for bRo5 compounds such as
Recently, some approaches have been put forward to address macrocyclic164,165 or degrader molecules,166 conformational
this issue.127−130 sampling has assisted in predicting permeability.
The prediction of degrader properties can be enhanced by Over the past decade, reports on atomistic simulations of the
molecular dynamics (MD) simulations, which have become an entire process of passive membrane transport have multiplied
important tool in computation-enabled drug discovery in the literature, as previously reviewed.167−169 For drug
assisting in the identification of ligand binding sites on protein molecules of different sizes, experimental permeability
surfaces,131−133 augmenting docking routines,134−136 and coefficients have been estimated based on the solubility−
predicting protein−ligand binding affinities.137−139 Compared diffusion model170−174 and, notably, distinct permeation
5414 [Link]
J. Chem. Inf. Model. 2023, 63, 5408−5432
Journal of Chemical Information and Modeling [Link]/jcim Perspective

pathways of (small-molecule) compounds have even been set of curated data,196 which are scarce for degrader molecules.
detected,175 highlighting the potential of novel simulation In this context, the ∼2-fold expansion of PROTAC-DB 2.0,197
algorithms, in combination with more realistic models of cell one of the primary repositories on structural and experimental
membranes,176 to advance our understanding of the cellular data on degraders, which has recently grown to over 3,200
uptake of drugs. Yet, despite these successes, all-atom entries at the time of this writing, mitigates this dearth of
simulations of translocation across membranes remain costly knowledge to some extent. However, the data are still sparse
to be applied at scale for relatively large molecules like given the complexity of the problem. We expect that the
degraders. To this end, low-dielectric continuum or implicit collection of information about degraders, including their
membrane models177−181 and (physics-based) mechanistic chemical structures, biological activities, and physicochemical
predictors, such as PerMM,182,183 COSMOperm,184 or other properties, will continue to grow over the next decade. To
tools,185 have been developed, which allow higher throughput, compensate the lack of available experimental data in the near
albeit at the cost of accuracy. future, data resampling techniques198,199 may be applied that,
The application of ML methods for the prediction of in combination with dynamic structural features readily
membrane permeability has also sharply increased in recent available through molecular simulations, could facilitate the
years. Some examples include neural networks, support-vector development of models for degrader classification.
machines, and random forest classifiers trained on molecular 3.2. Conformational Sampling of Degrader-Induced
structures and physicochemical properties of small and large Complexes Plays an Important Role in TPD Modeling.
compounds.186−189 To predict degrader permeability, Poonga- The most-studied step in the TPD process by means of
vanam et al.190 have tested several binary classifiers using molecular modeling is the degradation-induced formation and
descriptors that represent molecular size, shape, and chemical conformational sampling of ternary complexes. As discussed,
functionalities. While predictions were good (accuracy >80%) structure-based biophysical experiments, in particular, X-ray
in some cases (e.g., VHL-recruiting degraders), the classifiers crystallography, have previously helped in the rational design
performed poorly on cereblon-recruiting degraders due to of new degraders. However, since ensembles of structures
imbalances in the corresponding training data. This empha- better aid in understanding structure−function relationships
sizes the need for more high-quality degrader data sets, than static structures do, the optimization process of potent
especially given the hundreds of E3 ligases in the human degraders can strongly benefit from accurate modeling of
genome, many of which have minimal TPD data. Rather than ternary structures, i.e., the prediction of distinct molecular
directly measuring permeability, which tends to be challenging interactions that promote complex formation, thus character-
for heterobifunctional degrader molecules, cellular target izing the underlying selectivity and cooperativity.
engagement assays, described in more detail in Section 2, are For a robust assessment of predicted ternary complexes, the
often deployed to assess whether a molecule gets into the cell structural flexibility of degraders along with the variability in
or not (although permeability is not measured directly through productive POI−ligase poses requires a comprehensive
this approach). sampling of possible conformations of all components. For
The examples presented above, as well as other references in example, it has been shown that the linker motif impacts the
the literature,66,154 highlight the two obvious difficulties in the ternary complex formation and degradation efficiency.200,201
prediction of degrader properties such as solubility and Furthermore, the notion of cooperativity calls for a
permeability. First, due to their relatively large and representative set of protein−protein assemblies to identify
heterobifunctional structure, degraders can adopt a multitude possible degrader effects. To this end, several prominent
of different conformations that, unlike small-molecule drugs, docking programs, often augmented with MD, have recently
can lead to very different molecular interactions, thus been employed for the design of selective de-
complicating traditional property-based drug design. Yet, it is graders.38,45,89,202,203
this structural flexibility that leads to the aforementioned Ternary complex docking has emerged as an extension of
degrader chameleonicity, which allows both solubility and traditional docking methods. Molecular docking is a standard
permeability. Conformational sampling via molecular dynam- tool for pose prediction and virtual screening in drug
ics, Monte Carlo, or other techniques can be used to capture discovery.204,205 In the context of degrader-induced TPD,
these effects and generate molecular descriptors of degraders, both protein−ligand206 and protein−protein207 docking may
which we believe to be important in degrader design. be applied to generate structures that are not experimentally
Molecular simulations can furnish conformational ensembles resolved. In fact, several protocols have been developed that
that help to derive structural attributes, such as hydrogen differ by the order in which degrader conformations are
bonding and solvent accessibility, in different environ- sampled, i.e., either after all components are docked into a
ments.191,192 The use of such 3D descriptors, in particular ternary complex, or only in a binary protein−degrader complex
for the prediction of solubility and permeability, is currently an before superposing the second protein, or even before inserting
active area of research.29,164,165,193,194 the degrader with its sampled conformation into a given
The second main issue in TPD modeling is that accurate protein−protein aggregate.208 While each docking protocol
prediction of degrader properties is currently strongly attempts to minimize the number of predicted structures with
hampered by the lack of experimental data. Although it had sterically clashing components, those sampling the POI−ligase
been argued before that the actual QSPR algorithms, rather interactions and the degrader conformations separately before
than the uncertainty in data measurements, may be responsible combining all parts into a ternary complex have become most
for inaccurate solubility predictions,195 it is universally popular due to their higher accuracy in recapitulating
accepted that limitations on training data impair prediction experimentally known or crystal-like structures.209−211 Typi-
accuracy, especially for ML-based approaches. Robust cally, repurposing recently developed automated rou-
predictive modeling of physicochemical traits, such as tines,212−214 a library of linker conformations is combined
solubility and permeability, requires a large and representative with the warhead and the E3 ligand motifs, that are possibly
5415 [Link]
J. Chem. Inf. Model. 2023, 63, 5408−5432
Journal of Chemical Information and Modeling [Link]/jcim Perspective

Figure 4. Conformational sampling from MD simulations of (a) a degrader molecule in solution, (b) a degrader−ligase binary complex, (c) a
POI−degrader−ligase ternary complex, and (d) a POI−degrader−CRL supramolecular assembly (not to scale). As explained in the main text,
simulations can be used to predict the properties of degraders, the affinity of binary complexes, dynamic ensembles of ternary complexes, and the
likelihood of ubiquitination in the context of the CRL.

docked to structures from the POI and the E3 ligase, thus demonstrating the value of advanced simulations for the study
generating ternary complex models, which are scored by an of the TPD process.
energy function to identify favorable combinations.28,40,45,89 In our opinion, MD simulations should be fully integrated in
The interactions within each protein are unlikely to change degrader design cycles (see Section 4) as a means to provide
significantly in the context of the ternary complex, simplifying mechanistic information as well as thermodynamic and kinetic
the search for high-scoring models. Although recent work parameters for the decision-making process. Figure 4 displays
argues for the simultaneous docking of all three components the structural variability of the degrader molecules and distinct
for higher accuracy,215 in some instances, the binding modes of degrader-induced complexes sampled by MD simulations.
the warhead to the target POI and that of the E3 ligand to the Despite the structural complexity of the molecular systems and
E3 ligase are known, allowing for constraints to be applied to the associated time scales of the processes involved in ternary
the ternary complex docking problem. Specifically, the use of complex formation and target ubiquitination, modern
such structural information on the ternary complex interface enhanced sampling algorithms, in combination with graphical
obtained from HDX experiments, as implied above, has been processing units (GPUs), offer substantial progress in the size
shown to significantly boost the performance of ternary and time scale of simulations that can be routinely performed.
docking protocols.107,108 Specifically, simulations can augment and greatly improve
Considering the relatively intricate problem of ternary the accuracy of ternary complex docking protocols, more so
complex formation, MD simulations hold great promise for than already acknowledged in the context of small-molecule
rationalizing structure−activity relationships among the differ- docking.224,225 MD applied to docked ternary complexes can
ent binding partners. To date, simulations of ternary complexes assess their quality and possibly simulate transient and
(meta)stable states. In contrast, applying ternary complex
have been applied for tasks such as analyzing interactions
docking after the simulation of adjacent POI−ligase pairs that
between the three binding partners in atomistic detail,216
lack a degrader, so-called “encounter complexes”, allows the
probing the binding cooperativity,217 scoring and ranking given
assembly of a ternary complex with distinct protein−protein
degraders,218 and assessing the stability of degrader variants,
interactions, which we find a very appealing strategy. From a
such as covalently binding degraders,219 macrocyclic de- statistical−mechanics viewpoint, the presence of a hetero-
graders,43 or such aimed at new diseases.220 In particular, bifunctional degrader between two proteins corresponds to a
molecular simulations have successfully augmented many constraint leading to sampling of a confined subspace within
biochemical and proteomics assays to inform on the selectivity the full encounter complex configurational space. Therefore,
of distinct degraders38,89,221,222 and also structural biophysical simulations of degrader-less encounter complexes exhibit the
techniques to elucidate binding site interactions.40,93,223 baseline interactions of a given POI−ligase pair, which are not
Recently, our research team has combined MD simulations accessible from simulations of the full ternary complex,
with X-ray crystallography, small-angle scattering, HDX, and providing knowledge that can be exploited for rational
ubiquitinomics experiments to explore the differential in DC50 degrader design.
values between three VHL-recruiting SMARCA2 degraders.108 Path-sampling strategies are employed to simulate non-
This rather comprehensive work provides an explanation for equilibrium phenomena, such as the degrader-induced
degradation efficiencies as primarily influenced by the stability assembly of proteins. These methods, also referred to as
and geometry of ternary complexes, particularly in the context importance-sampling techniques, usually require a predefined
of the entire Cullin−RING ligase (CRL). Based on collective variable along which phase space is sampled,
simulations, we found that productive ternary complexes, i.e., accomplished by the addition of biasing potentials as in
such with a high ubiquitination probability, do not coincide umbrella sampling226 or steered or accelerated MD227,228 or
with the crystal structures, which presumably are dominated by similar methods, that apply the potentials in an adaptive
crystal contacts, but rather with the global energy minima, thus fashion.229−231 One such method, that has become popular in
5416 [Link]
J. Chem. Inf. Model. 2023, 63, 5408−5432
Journal of Chemical Information and Modeling [Link]/jcim Perspective

computational drug discovery, is metadynamics232−234 and its combined with the enhanced simulations, possibly even as
many variants that converge faster,235−239 which, in our study, iterative reweighting schemes.254,255 These approaches often
were adequate for the simulation of large structural changes in project the dynamical evolution of the system in space and
the RING-finger E3 multiprotein ligase complex connected to time onto a model, as in master equation representations,256
SMARCA2 by a degrader.108 Simulations of large conforma- capturing the conformational dynamics of the system and
tional changes of the ubiquitination system are instrumental for enabling long-time scale predictions. A popular concept is the
modeling the TPD process and, in our opinion, enhanced use of Markov State Models (MSMs), which, over the past
sampling methods are key in this regard. For instance, decade, have been applied to a large number of biomolecular
simulations that mechanistically describe the ubiquitination systems.257−260 MSMs provide one of the best ways of coarse-
process can model the presentation of the target substrate, i.e., graining the dynamics of the ternary complex ensemble and
its proximity and orientation, with respect to the ubiquitin- identifying high-probability conformational states that can be
carrying enzyme, thus supporting the assessment of degrader correlated to ubiquitination scores and other properties. This
efficacies or the degree of (poly)ubiquitination of distinct sites. strategy can help identify specific residues that may be of
In contrast to the methods described above, rigorous path- particular importance to transitions among different states and
sampling schemes introduce less bias into the dynamics and also quantify the effect of degraders on the dynamics of the
thus do not require any additional assumptions for the ternary system.
calculation of rate constants of the simulated rare event. These Finally, we should mention the recent rise in de novo protein
include, among other methods, transition path sampling240,241 structure prediction, based on contacts and sequence
or dynamic importance sampling,242 in which complete information from known structures,261−265 as an avenue to
transition paths are iteratively refined or reweighted, or predict conformations of productive ternary complexes. While
strategies that construct paths by simulating many short these approaches are promising, their application to larger
trajectory segments, such as forward flux sampling,243 mile- protein aggregates, such as ternary complexes, is more
stoning,244 weighted ensemble,245,246 or innovative combina- complicated and has not been reported yet. Moreover, in
tions thereof.247 In the context of ternary complex formation contrast to many simulation methods, they lack information
and ubiquitination, we consider these simulation techniques to about the assembly and structural variability of ternary
be essential, as they yield kinetic and thermodynamic complexes. Nevertheless, we consider that the generative
information on binding events, thus complementing experi- modeling of degrader-induced ternary complexes will become
ments like SPR and ITC. Importantly, the simulation of an important analysis tool for TPD modeling as more
multiple (weighted) binding pathways can reveal the impact of knowledge on the determinants of productive structures is
degrader molecules on protein−protein association. By being accumulated.
estimating transition rates, these simulations can reveal which The methods outlined above highlight the significance of
pose of a ternary complex is most favored and thus measure advanced molecular simulations, not only in understanding the
the likelihood of distinct sites to be ubiquitinated. In fact, we fundamentals of TPD but also as a routine application tool for
have shown the capacity of weighted ensemble simulations to degrader design. As we describe in more detail in Section 4,
predict ternary complex binding pathways and reproduce computation-enabled degrader design relies on the orchestra-
binding rate constants that are in line with experiments.108 tion of automated procedures to efficiently scan individual
Notably, the simulations gave more accurate results, degrader candidates. The simulation approaches presented
particularly in comparison to ternary complex docking, when here provide a set of tools that are very well-suited for this task.
the collective variable included information from HDX 3.3. Mathematical Models Can Help Determine
experiments using an affordable amount of resources. Properties to Optimize Degradation. There is a complex
Generalized ensemble techniques, such as parallel temper- relationship within the in cellulo and in vivo environments
ing,248,249 are orthogonal to path sampling in that the between the aforementioned steps in the TPD process.
simulation is enhanced in a path-independent manner, for Complementary to molecular modeling, mathematical frame-
instance by raising the temperature. We have shown that such works furnish a representation of the TPD process at larger
an approach is beneficial for the exploration of ternary complex time and length scales. In addition to the relationships across
structures and dynamics, producing an accurate description of solubility, permeability, binding, ubiquitination, and degrada-
conformational states upon projection onto a free energy tion, there are subtleties associated with the stability of the
surface that aids in identifying metastable states within the ternary complex, which is influenced by the interactions
dynamic ternary system.108 Specifically, solute scaling (or between the degrader molecule, the POI, and the E3 ligase.
Hamiltonian replica-exchange MD, HREMD)250 and flexible The resulting positive or negative binding cooperativity during
tempering (REFT),251 in which the Hamiltonian of the whole ternary complex formation can influence the overall degrada-
solute or only parts of it are scaled, are promising alternatives tion efficiency. However, the impact of cooperativity on
to sample the structures of ternary aggregates as they are more degradation, mediated by a heterobifunctional molecule,
efficient than the traditional temperature REMD. A further remains an open question. In a recent study by some of the
option to efficiently explore the free energy landscape of authors of this Perspective, a pharmacodynamic model was
ternary complexes is the coupling of REMD to a reservoir of developed to describe the kinetics in the TPD process and it
different conformations,252,253 that could stem from ternary was used to explore the role of cooperativity in ternary
complex docking, virtually leading to the simulated annealing complex formation and POI degradation.266 The model
of a docked ensemble. established a quantitative relationship between the stability of
The methods described can simulate degrader binding, the ternary complex and degradation efficiency by examining
ternary complex formation, and conformational changes the effect of the complex stability on the rate of catalytic
(Figure 4). Still, considering the size and the flexibility of the turnover. Additionally, the authors devised a statistical
ternary systems, sophisticated analytical treatments need to be inference model to determine cooperativity in intracellular
5417 [Link]
J. Chem. Inf. Model. 2023, 63, 5408−5432
Journal of Chemical Information and Modeling [Link]/jcim Perspective

ternary complex formation using cellular assay data. The work 4.1. A Design Strategy Based on the Integration of
was validated by quantifying changes in cooperativity due to Computational Tools. While both the warhead and the E3
site-directed mutagenesis at the POI−ligase interface of the ligand motifs within a heterobifunctional degrader molecule
SMARCA2-ACBI1-VHL ternary complex. This pharmacody- are essential for ternary complex induction, a multitude of
namic model is an example of a quantitative framework to studies have demonstrated that potent and efficacious
dissect the intricate degrader-mediated TPD process, which degradation of a target POI is also dependent upon the
could inform the rational design of effective heterobifunctional conjugation vector and, as noted above, the chemical structure
degraders. By contextualizing these findings with the of the connecting linker motif.89,271,272 In this vein, many
experimental techniques described above, it should be possible computational tools for degrader design focus on linker
to provide a more comprehensive understanding of the factors conjugation and optimization. Biophysical and structural
influencing the success of TPD-based drug discovery and studies have revealed critical insights into how linkers influence
therefore a more rational approach to optimizing degrader the positioning of the POI in relation to the ligase, either
properties for TPD. positively or negatively regulating its ubiquitination.28,45 These
In a related work, researchers developed an extensive studies and biochemical investigations alike have also shown
modeling framework to analyze experimental data for three how alterations in chemical linkers can facilitate cooperativity
primary objectives: (1) evaluate degraders using precise and stability of ternary complexes, often leading to improved
degradation metrics, (2) optimize crucial compound parame- and more sustained degradation.28,37,40 Structural studies and
ters, and (3) link degradation to subsequent pharmacodynamic the advancement of molecular modeling of larger protein
effects.267 The proposed framework introduces several novel complexes have guided degrader linker design for targets and
features: (1) a mechanistic model to fit the hook effect ligases with prior information available.23,273,274 Therefore,
observed in degrader concentration−degradation profiles, (2) combinatorial approaches for degrader design typically use a
quantification of the role of target occupancy in the mechanism small number of warheads and E3 ligands and consider their
of action, and (3) disentangling the effects of target associated attachment points and linker libraries. A good
degradation and inhibition on the overall pharmacodynamic source of linkers is the patent literature on degraders,
response. The authors demonstrate the applicability of this commercially available linker libraries, and the aforementioned
approach by applying these three models to analyze exemplary publicly available PROTAC-DB.197
data from multiple compounds, projects, and targets. The A good degrader must be synthetically feasible, chemically
framework enables researchers to tailor their experimental stable, and bioavailable. Thus, similar to small-molecule drug
work, leading to a deeper understanding of their results and design cycles, any computational prediction on degrader
ultimately facilitating a more successful degrader discovery. properties should be followed by decision making in
Although the focus of this work was on in vitro pharmacology collaboration with synthetic and medicinal chemists. Consid-
experiments, the key implications for in vivo studies are also ering the many different types of linker motifs recently
discussed. suggested in the literature (for a good overview, see Figure 1
Along the same lines, a general model for ternary complex by Desantis et al.275), expert opinion is imperative for degrader
catalysis has been developed within a framework familiar to design. For instance, synthetic accessibility constraints should
classical enzyme theory.268 The authors adapted a strategy be employed, and synthetic handles defined a priori to ensure
within Michaelis and Menten’s original publication (integra- computational predictions can be validated experimentally.
tion of the velocity equation) to solve for the maximum An important step in degrader design is certainly the use of
velocity (Vmax).269 These equations are straightforward to physicochemical and ADMET property predictions to filter the
implement and enable estimation of time scales that are number of compounds being considered. However, as
consistent with a wide range of published literature values. discussed above, many of these methods are currently being
Additionally, the authors validated these equations with improved for degraders and will play a more significant role in
thermodynamic and kinetic databases and built an interactive the near future. For the prediction of permeability, we found
web tool that enables researchers to graphically simulate their that running MD simulations for 0.5 μs to obtain ensembles of
own systems. Other reviews have discussed the current state structures and thus distributions of quantities, such as the
and future directions of TPD drug discovery in the context of linker end-to-end distance, provides more accurate results than
building a quantitative relationship between loss of protein applying predictors to a single structure. Also, based on our
target and in vivo activity,270 where mechanistic PK−PD experience, the prediction of Log P values with (physics-based)
models are highlighted with the aim to improve the translation mechanistic tools, such as PerMM,182,183 is suitable to provide
from the preclinical to clinical space. a correct trend among multiple degraders.
The fastest way to obtain a ternary structure of a given
4. COMPUTATION-ENABLED DEGRADER DESIGN POI−ligase pair connected by a degrader candidate molecule is
Early in degrader discovery programs, when experimental via ternary complex docking. As we described earlier, the actual
information is sparse, computational modeling can be protein conformations are not changing significantly during
leveraged to design and prioritize molecules for synthesis. this procedure; therefore, the key challenge in ternary complex
The process of designing degraders includes multiple complex docking entails computing the orientation of the two proteins
steps incorporating computational tools, which have the relative to each other. We use ternary complex docking with
advantage of prospectively generating degraders and comput- two major modifications. First, rather than using rigid protein−
ing quantitative rankings that can inform decision-making in protein orientations generated by protein−protein docking, we
discovery projects. To this end, we have developed a workflow use conformations coming from MD simulations of the
that comprises the modeling and simulation methods encounter complexes (augmented by Markov State Modeling,
presented above to guide the degrader design and assess the as described below) as the starting point for docking the
suitability of degrader candidates. degrader candidate to form a ternary complex, thus, capturing
5418 [Link]
J. Chem. Inf. Model. 2023, 63, 5408−5432
Journal of Chemical Information and Modeling [Link]/jcim Perspective

Figure 5. A simulation workflow to assess the quality of the degraders (illustrated here for the bromodomain of SMARCA2 as the target POI and
VHL as the ligase). (a) The formation of ternary complexes is achieved by WE simulations, generating a variety of different ternary conformations.
(b) Enhanced HREMD sampling helps explore ternary complex structures that may significantly differ from each other, producing a converged free
energy surface. The star indicates the conformation of the crystal structure. (c) Analysis of ternary complex conformations in the context of the full
CRL assembly, exhaustively sampled by meta-eABF, allows the assessment of the ubiquitination probability of the POI.

important baseline interactions between the POI and the E3 Components Analysis (tICA),276 thus extracting a set of
ligase in order to achieve accurate ternary models. Second, POI−ligase encounter complexes that, based on the given
those ternary complexes that minimize clashes, preserve simulations, correspond to their metastable states. The
warhead and E3 ligand binding modes, and, if known, look significance of this simulation-driven approach is that
similar to ternary complexes of active degraders are scored degraders can be developed to optimize the balance between
highly, whereas those models that look similar to ternary enthalpic and entropic contributions to ternary complex
complexes formed by known nondegraders are penalized. This stability, for instance, such that they stabilize a given state
orientational sampling thus defines what linker geometries are with favorable interactions at the protein−protein interface or
tolerated and preferred for each ternary complex. rather facilitate the conformational change between two
Since a large number of different protein−protein poses or metastable states. In our MSM protocol, we use K-means (or
orientations may get sampled, the encounter complex K-centers) clustering based on the number of contacts (i.e.,
simulations can produce a vast amount of data that require heavy atoms within 5 Å) formed between the interprotein
reduction. We apply Markov State Modeling (MSM) to the residues during the MD simulations. Importantly, despite the
simulation trajectories and characterize their slowest-relaxing fact that the protein−protein simulation is more expensive
degrees of freedom with a Time−structure Independent than the rigid protein−protein docking, this simulation has to
5419 [Link]
J. Chem. Inf. Model. 2023, 63, 5408−5432
Journal of Chemical Information and Modeling [Link]/jcim Perspective

be performed only once for a given POI−ligase pair, resulting distance profiles should be evaluated for each choice of
in a set of encounter complexes that can be reused for ternary simulation trajectory and model of the CRL.
complex docking of a myriad of degrader candidates. We augment our modeling and simulation workflow for
We point out that the ensemble of simulated encounter degrader design with a random forest classifier trained on
complexes allows the direct design of degraders based on the known degraders and nondegraders of the target protein under
baseline interactions mentioned above, which are, in practice, consideration. Nondegraders are heterobifunctional molecules
preferred protein−protein orientations or interfaces. This that have been experimentally verified to have very little or no
strategy essentially incorporates the design process into the degradation activity, which may be due to a variety of reasons.
ternary complex docking, virtually using the POI−ligase If available, experimental degradation data (e.g., Dmax) should
encounter structures as “templates” to draft a new degrader be included in the model in addition to structural and
molecule. physicochemical descriptors of the compound, such as the
Generating ternary complexes with candidate molecules molecular weight and estimations of polarity and permeability,
should be followed up by MD simulations to examine their as well as features extracted from MD simulations of a fully
relative stability, preferably in the context of a supramolecular solvated degrader that characterize its flexibility, like the
assembly to predict more accurately the ubiquitination (normalized) linker end-to-end distance or its gyration radius,
probability of lysines in the POI. For estimating stability of and also such obtained from the ternary complex (HREMD)
the complexes, we recommend using proxy metrics for stability simulations, for instance, interface contacts and the lysine−
such as RMSD, RMSF, and the solvent-accessibility of lysine ubiquitin distance distributions. We apply a principal
side chains over time. component analysis to reduce the dimensionality in feature
More rigorously, as illustrated in Figure 5, we have space and eliminate strong correlations among features. To
implemented a multistep simulation routine to assess the train such a model for a relatively accurate classification of
quality of degraders, that is, their ability to form a stable degrader candidates, we suggest to have a (well-balanced) set
ternary complex with high ubiquitination probability. These of at least 50 degrader/nondegrader molecules, to use two-
simulations are applied to new degrader designs, but also to thirds for training, and to cross-validate the ML model.
experimentally confirmed degraders and nondegrading hetero- Identifying the most predictive features tells us what input
bifunctional molecules, that, as discussed below, would then properties are crucial for degradation of the POI studied,
serve as input classes for a classifier to categorize a candidate which can strongly support the design process.
molecule as a degrader or a nondegrader and also to guide the Our initial studies on designs of SMARCA2 degraders
(described in more detail in Section 4.2) have revealed that the
design of new degraders.
inclusion of all three feature categories, i.e., properties derived
To drive the formation of ternary complexes (see Figure 5a),
from the molecular structure, such obtained from fully solvated
we apply weighted ensemble (WE) simulations starting with
degrader simulations, and such from the ternary complex
the well-separated POI and the ligase−degrader binary
simulations, leads to classifiers with predictive ability of >80%
complex (or vice versa) and using as a collective variable the
accuracy, while those trained on only one of the mentioned
RMSD of only the warhead (or E3 ligand) to a corresponding
categories are still sufficiently accurate (70−75%). Thus, to be
bound structure of the other binary complex. The “Resampling operationally prudent, we suggest applying the random forest
of Ensembles by Variation Optimization” (REVO)277 is the classifier to only the physicochemical and structural features of
WE algorithm of choice, as it yields ternary complexes with a degrader designs in order to filter a smaller subset before
variety of different conformations through the iterative running the more expensive simulation workflow and applying
optimization of “trajectory variation”. the ML model again, including features extracted from
Then, HREMD simulations are applied to bound ternary simulations.
complexes using 20 replicas with an effective maximum Using 64 GPUs and about 500 CPUs, we were able to
temperature of 425 K, generating a more detailed map of the operate our design pipeline in 1 week, which is the appropriate
ensemble of ternary complexes (Figure 5b). We suggest to use time scale for decision making in a discovery program. The
a projection of the sampled conformations that epitomizes the random forest models require only a few seconds for thousands
interface, such as the principal components of site−site or millions of virtual designs. Most of the computation is spent
distance distributions.108 These conformational landscapes on docking of 500−1000 degrader candidates and MD
shed light on the stability of a ternary complex, and simulations for the top 50 degrader compounds.
importantly, they can inform the docking scoring functions Although our design strategy features multiple modeling and
for the next round of designs. simulation techniques, it must be complemented by experi-
Finally, to estimate a degrader’s impact on the ubiquitination ments. Promising degrader candidates, i.e., those that have
of a POI, we superimpose the ternary complex structures formed stable ternary complexes with a high ubiquitination
obtained from HREMD on the full ubiquitin-bound Cullin− probability and that have been labeled as a degrader by the ML
RING ligase (CRL) complex, which has been sampled classifier, must be tested by biochemical assays, e.g., by ALPHA
exhaustively before between open and closed conformations or HiBiT, to attest their ability to form ternary complexes or
by the metadynamics variant meta-eABF238,239 (Figure 5c). degrade the target. This experimental feedback is instrumental
The distribution of distances between ubiquitin and different in degrader design cycles, which, as mentioned earlier, are
lysine residues on the target protein’s surface, computed over similar to any drug design cycle in that experimental expertise
all frames, can be used as a scoring function regarding the is necessary for decision making.
propensity of ubiquitination induced by the simulated 4.2. Application to a Degrader Design Project. We
degrader. An alternative procedure that is also robust for demonstrate the impact of computation by briefly presenting
predicting ubiquitination scores is to directly simulate the some results from our degrader discovery project, in which we
ternary complex in the CRL, rather than superimposing. Lysine designed heterobifunctional degraders for SMARCA2-VHL
5420 [Link]
J. Chem. Inf. Model. 2023, 63, 5408−5432
Journal of Chemical Information and Modeling [Link]/jcim Perspective

Figure 6. Linker designs. ACBI1 is the potent and cooperative SMARCA2 degrader optimized previously.28 Linkers 1−5 (orange frame) were
shown to yield high SMARCA2 ubiquitination probabilities in simulations. Linker 6 (purple frame) was designed based on a metastable
SMARCA2-VHL encounter complex.

with a novel linker motif supported by our simulation degradation potency of these designs, and one of them had
workflow. Figure 6 lists all of our designs in this study Dmax = 54% and DC50 = 174 nM, which is still acceptable in
(Linkers 1−6) along with ACBI1, a previously optimized terms of degradation activity (see experimental results in
VHL-recruiting degrader of SMARCA2.28 Informed by the Figure 7). By contrast, 34 designs that incorporated typical
random forest classifier, which was trained and validated on alkyl and PEG linkers spanning 4 to ∼20 atoms as well as
about 100 known compounds (model accuracy of 83%), the known linker moieties, such as triazoles and phenyl rings (o, m,
designs generally involve rigid linkers that contain (aromatic) or p-substituted to provide different geometries), only
ring structures with low molecular weight. Hence, five linkers produced two active degraders.
were designed based on pyridine, pyrimidine, dioxolane, and Another demonstration of the impact of computation is the
azetidine heterocycles (Linkers 1−5 in Figure 6) with the design of a SMARCA2-VHL degrader with a novel protein−
objective of stabilizing the ternary complexes and improving protein interface. Briefly, about 150 SMARCA2-VHL con-
the associated ubiquitination probability of lysine residues on formations were generated from rigid protein−protein
the surface of SMARCA2. As shown in Table 2, three out of docking. 150 independent MD simulations of 3 μs were run
starting at each one of these docked conformations to capture
Table 2. Summary of the Experimental Results on the baseline interactions between SMARCA2 and VHL.
Degradation Activity of the Individual Linker Designs Markov state modeling revealed three main metastable states:
Design DC50 (nM) Dmax (%) AlphaLISA (% of control)
one of the metastable states recapitulated the known crystal
structures of SMARCA2-VHL ternary complexes (PDB IDs
ACBI1 37 94 100
6HAX, 6HAY, 74SE); another metastable state had a
Linker 1 80 86 69
completely different protein−protein interface, in which
Linker 2 n/a 38 52
SMARCA2 was indeed rotated by about 180° compared to
Linker 3 27 97 11
the known crystal structures as illustrated in Figure 8. Using
Linker 4 43 97 20
Linker 5 174 54 20
that latter state for structure-based design, we produced a
Linker 6 67 90 78
relatively short linker motif including a pyrrolidine group and
with attachment points different from those in the previously
resolved crystal structures or in any of the past designs (Linker
these five designs were found to have Dmax > 85% and DC50 ≤ 6 in Figure 6). Our simulation workflow (see Figure 5)
80 nM in the corresponding experiments, validating the high revealed that ternary complexes with this degrader candidate

Figure 7. Experimental confirmation of degrader designs. (a) Fitted curves from AlphaLISA assays measuring ternary complex formation. (b)
Degradation profiles from HiBiT assays.

5421 [Link]
J. Chem. Inf. Model. 2023, 63, 5408−5432
Journal of Chemical Information and Modeling [Link]/jcim Perspective

Figure 8. Comparison of ternary complexes with designed linkers. (a) Superposition of ternary complex crystal structures with Linker 6 (red
SMARCA2 structure), designed in our study based on the simulation of SMARCA2-VHL encounter complexes, and with the previously designed
ACBI1 degrader (green SMARCA2 structure).28 The structures are superimposed with the VHL ligases aligned. The transparent SMARCA2
structure is the model protein−protein encounter complex with which Linker 6 was designed. The contours indicate the electron densities of the
corresponding warhead structures. (b) A free-energy landscape of the SMARCA2-VHL encounter complex conformations obtained from molecular
simulations. tIC0 and tIC1 are the two slowest-relaxing degrees of freedom based on linear combinations of interface residue contacts between the
two proteins observed during simulation as described elsewhere.108

were relatively stable and SMARCA2 very likely to be We envision that in the future, a variety of different
ubiquitinated. computational protocols would be applied for TPD analysis
As displayed in Figure 8a, the crystal structure of the ternary and degrader design. Depending on the POI−ligase system
complex with Linker 6 (red structure) is similar to the and the availability of data on known active degraders, we
corresponding simulated structure (transparent structure), anticipate that the focus might lie on predicting only distinct
based on which the degrader was designed, thus validating features that could complement experiments. For instance,
this approach of a simulation-driven structure-based degrader these may be the likelihood of certain lysines being
design. Furthermore, Figure 8a confirms that this degrader ubiquitinated in ternary complexes or certain regions on the
induces an obviously different conformation compared with protein surfaces being solvent-exposed, which can be assessed
the previously known ternary complex crystal structure with by the simulation methods discussed. Also, if the rapid
ACBI1 (green structure). Despite the distinctly different screening of designs is required for decision making, ternary
conformations, the interface−RMSD of these two ternary complex docking, followed by short MD simulations, will
complexes is 2.6 Å and about 40% of the same interface atoms remain the primary approach. However, as our research results
demonstrate, a rigorous strategy to examine degrader designs
are in contact between the two corresponding crystal
and explore protein−protein interactions is often necessary
structures. Figure 8b visualizes a free energy landscape of the
and, evidently, achievable.
SMARCA2-VHL encounter complex simulations that capture
all baseline interactions of this POI−ligase pair. Obviously, the
5. FUTURE DIRECTIONS
ternary complex crystal structures with ACBI1 and Linker 6
occupy different metastable states. The degrader candidate This is a unique time in the field of targeted protein
based on Linker 6 was classified as a degrader by our ML degradation (TPD). After decades of academic research, we
model, which has been confirmed experimentally with Dmax = are now seeing biotechnology and pharmaceutical companies
90% and DC50 = 67 nM. bring TPD molecules to the clinic. While it appears that
The results presented for the SMARCA2-VHL degrader computational tools were used in the design of some clinical-
design underscore the suitability and the benefit of the stage molecules, their impact was, in fact, minimal. However,
there is a growing number of examples where computational
modeling and simulation strategies we developed. In particular,
approaches are being used to address some of the great
the prediction of a favorable, previously unknown, protein−
challenges in TPD drug discovery.
protein interface, which served as a template to design an As discussed in this Perspective, there is a diverse array of
active degrader, is, in our opinion, a remarkable achievement, computational tools contributing to the TPD field. In some
showing the potential of a simulation-driven protocol and its cases, traditional tools can be easily repurposed from small
complementarity to experiments. Importantly, we believe that molecule applications to TPD. For example, docking and
our research has highlighted how computational methods, screening of warheads and E3 ligands are analogous to small
similar to experiments, permit us to approach the task of molecules and can therefore be used directly. Similarly, many
degrader design from different angles. We have combined of the underlying methods to predict properties are the same
results from simulations that examine POI−ligase interactions (e.g., QSAR and machine learning based on experimental
with such data that assess the stability of ternary complexes data), although the data are quite sparse and the chemical
and its ubiquitination probability in the context of a space is much larger for many TPD molecules, making the
supramolecular protein aggregate, showing how the integration existing models less predictive. In other cases, traditional tools
of multiple molecular simulation (and docking) methods can require additional training and parameter tuning to improve
support degrader design. results for TPD, such as protein−protein docking algorithms,
5422 [Link]
J. Chem. Inf. Model. 2023, 63, 5408−5432
Journal of Chemical Information and Modeling [Link]/jcim Perspective

where constraints related to the binding sites and warhead/ MD trajectories of ternary complexes as graphs that are
ligand attachment points can significantly reduce the search transformed and used for the training of feed-forward networks
space, thereby improving both the speed and accuracy of the to predict the functional form defined by the raw HiBiT data
algorithms. over time, enabling the calculation of pharmacological
Perhaps the most interesting, innovative, and impactful is the constants such as DC50, Dmax, and the degradation rate.
growing number of approaches to simulate the dynamic These approaches and other experimental data can be used in
behavior of the ternary complex. The significance of these many ways to improve our understanding of the TPD process,
methods lies in the fact that the formation of a ternary such as building mathematical models to connect quantities
structure is a necessary step in the TPD process and, that can be more readily computed (e.g., permeability, affinity,
furthermore, that the non-native protein−protein interactions and stability of the ternary complex) to important downstream
seem to be “floppier” than many endogenous protein−protein processes (e.g., degradation efficiency).266
complexes. Indeed, biology is in constant motion, and not The integration of the computational and experimental
surprisingly, TPD follows the same paradigm. The importance methods introduced in this Perspective is the key to success in
of understanding the induced ternary structures stems from the TPD research. In our opinion, which is based on the state-of-
criticality of this step in the degradation process, where the the-art methods and the promising results presented here,
ubiquitination mechanism requires not just binding but also there are three main avenues of combination: first,
forming the correct orientation of the protein of interest in experimentally generated data on degrader physicochemical,
relationship to the rest of the supramolecular assembly that is binding, and activity properties support predictive models;
responsible for the ubiquitination. This is supported by a second, structural proteomics and biophysics enhance docking
growing body of evidence, both computational and exper- and simulation procedures; and third, monitoring the degree of
imental. ternary complex formation and target degradation is necessary
In addition to dynamics and conformational variability in the for computation-enabled degrader design cycles. This level of
ternary complex, early works that simulate the full supra- interdependence between experiments and computation calls
molecular complex (e.g., the Cullin−RING ligase) leading to for a coordinated community-wide effort. However, to bring to
ubiquitination of the protein of interest have yielded promising fruition the promise of proteasomal degradation as a
results and significant insights into the TPD process. These therapeutic modality, collaborations must go beyond simple
simulations involve hundreds of thousands of atoms and information exchange; rather, they must comprise multi-
therefore require significant computational time and resources; disciplinary research teams dedicated to exploring the different
fortunately, these simulations will become more accessible as facets of TPD. Specifically, for the design of degrader
computer hardware continues to grow in power, efficiency, and candidates, as we outlined above, simulation or data mining
affordability. Still, for years to come, brute force simulations of results must be translated and implemented upon careful
meaningful time scales for TPD will be out of reach for most deliberation with synthetic and medicinal chemists. While such
researchers and restricted to specialized hardware like the teamwork has always been required in drug discovery settings,
Anton supercomputer278 or massively distributed systems like it is all too often neglected. The intriguing task of developing
[email protected] Fortunately, enhanced sampling algorithms heterobifunctional degrader molecules with high specificity and
can greatly accelerate simulations, especially when there is potency shows quite plainly how the cooperation between
knowledge about the collective variable (CV) of interest. In the both modelers and experimentalists could lead to sustained
case of the Cullin−RING Ligase (CRL), there is a growing productivity.
body of biophysical and structural biology data that facilitates While much progress has been made in the TPD field, more
the elucidation of practical CVs, as described in this work is needed in our quest to more efficiently design more
Perspective. effective TPD therapeutics. First, traditional QSAR models
Aside from the computational approaches discussed here, we generally perform poorly for large heterobifunctional degrader
are seeing many other technology advances related to TPD. molecules. A combination of more experimental data coupled
Structural biology techniques like cryogenic electron micros- with improved QSAR modeling approaches (perhaps account-
copy (cryo-EM) have enabled atomic-resolution structures for ing for 3D or even dynamic information) is likely required.
supramolecular protein assemblies like the CRL. Still, structure Improved docking and structure prediction methods would
(even with dynamics) is insufficient to design drugs. When also be of significant value. Ternary complex docking is a
developing degrader compounds, it is important to optimize relatively new problem, and most early applications have
additional properties, such as cellular permeability and affinity. involved retrofitting existing tools. It is likely that new
Approaches such as the NanoBRET target engagement assay algorithms, built from the ground up, will be better at solving
provide a quantification of interactions in live cells, which this specific problem. Finally, molecular design tools for the
encapsulates permeability, affinity, and residence time. Mass TPD are greatly needed. Repurposing tools like DeLinker213 in
spectrometry-based chemoproteomics is also enabling TPD our experience has provided some value, but there are still
discovery efforts, from screening for chemical scaffolds that significant gaps in terms of designing degraders that can be
bind to proteins in their native environment, including post- readily synthesized and have good ADMET properties.
translational modifications, to characterizing the time-depend- When developing new tools, the broad direction of the TPD
ent degradation process in the cell. Additionally, as more high- field should be considered. Most early work focused on
content experimental information becomes available, such as degradation through inducing proximity to enable ubiquitina-
live-cell kinetic data from technologies like HiBiT, the ability tion. In this context, covalently binding degraders are
to predict and improve degradation profiles becomes more considered to be more efficacious.219,280 Also, many of the
amenable to advanced ML algorithms such as 4D equivariant approaches, that we presented in this Perspective on
graph transformer representations of simulated ternary heterobifunctional degraders, can be applied to the character-
complex molecular dynamics. This specific approach encodes ization of molecular glues, which, however, presents its own set
5423 [Link]
J. Chem. Inf. Model. 2023, 63, 5408−5432
Journal of Chemical Information and Modeling [Link]/jcim Perspective

of challenges, such as the screening for suitable chemical

matter or the identification of an appropriate E3 ligase for a
■ ACKNOWLEDGMENTS
This Perspective is shaped by the research performed in the
given target POI. Moreover, we are now seeing the surfacing of
Advanced Simulation group of Silicon Therapeutics and
many other modes of degradation, such as lysosome-targeting
chimeras (LYTACs281), macroautophagy degradation targeting Roivant Discovery in 2021−2022. We thank all other team
chimeras (MADTACs282), autophagy-targeting chimeras members for their invaluable input (listed in alphabetical order
(AUTOTACs283), deubiquitinase-targeting chimeras (DUB- of last names): Simon Boothroyd, Mihir Date, Taras
TACs 284 ), and chaperone-mediated protein degraders Dauzhenka, Tom Dixon, István Kolossváry, Samuel Lotz,
(CHAMPs285). Additionally, induced proximity is being Derek MacPherson, Zachary McDargh, Tushar Modi,
leveraged for nondegradation applications, such as phosphor- Benjamin Mueller, Rajat Pal, Sharon Shechter, Utsab Shrestha,
ylation-inducing chimeric small molecules (PHICS286), where Fabio Trovato, Rafal Wiewiora, and Wenchang Zhou. We
a small molecule brings a kinase into proximity with a protein would also like to thank Alexander Tropsha for thoughtful
of interest to phosphorylate the target protein, which is not discussions during the preparation of this Perspective.
otherwise a substrate for the kinase.
We see a bright future for the field of TPD, and induced
proximity approaches more broadly, with encouraging clinical
data emerging and a growing number of innovative companies
■ GLOSSARY
Targeted Protein Degradation (TPD): A biomedical
in the preclincial stage. We envision continued improvements research strategy that aims to selectively eliminate specific
across many different computational and experimental proteins from cells for therapeutic applications, often using
methods that will contribute to a rich ecosystem that heterobifunctional degraders or molecular glues.
companies will leverage to build integrated workflows that Heterobifunctional: A molecule or agent with different
solve some of the most critical challenges in the field. structural motifs (or functional groups), enabling simulta-
Ultimately, with the confluence of advances in different neous interaction with multiple biological targets.
methods, we expect to see the emergence of next-generation Proteasome: A protein complex that degrades unneeded
workflows that will show demonstrable advantages over or damaged proteins by proteolysis, a chemical reaction that
breaks peptide bonds.
approaches that do not leverage computation. The greatest
Degrader: A heterobifunctional molecule designed to
successes will come from the computational efforts that are
induce TPD, bridging a specific protein of interest with an
tightly integrated with advanced experimental approaches in an
E3 ubiquitin ligase to trigger proteasomal degradation
iterative fashion. We hope that our Perspective will spark some (often referred to as a PROteolysis TArgeting Chimera or
ideas and contribute to the future of the exciting field of PROTAC).
targeted protein degradation. E3 ligase: An enzyme that plays a crucial role in the

■ AUTHOR INFORMATION
Corresponding Author
ubiquitin−proteasome system, catalyzing the transfer of
ubiquitin to specific target proteins, thereby marking them
for degradation.
Protein of interest (POI): The target protein to be
Jesus A Izaguirre − Differentiated Therapeutics, San Diego,
eliminated through the TPD process.
California 92056, United States; Atommap Corporation,
Warhead: The structural motif of a heterobifunctional
New York, New York 10013, United States; [Link]/
degrader that is designed to bind a specific protein of
0000-0002-4687-4884; Email: jesus@[Link]
interest and to induce its degradation.
Authors E3 ligand: A small-molecule ligand of the E3 ligase that
Barmak Mostofian − OpenEye, Cadence Molecular Sciences, facilitates its interaction with a target protein.
Boston, Massachusetts 02114, United States; [Link]/ Linker: A structural motif that connects the warhead to
0000-0003-0568-9866 the E3 ligand within a heterobifunctional degrader,
Holli-Joi Martin − Laboratory for Molecular Modeling, facilitating proximity and interaction for targeted protein
Division of Chemical Biology and Medicinal Chemistry, degradation.
Eshelman School of Pharmacy, University of North Carolina, Hook effect: Also known as the prozone effect, it refers to
Chapel Hill, North Carolina 27599, United States the phenomenon in which the efficacy of bivalent molecules,
Asghar Razavi − ENKO Chem, Inc, Mystic, Connecticut like heterobifunctional degraders, decreases at high concen-
06355, United States trations due to the increased formation of unproductive
Shivam Patel − Psivant Therapeutics, Boston, Massachusetts (binary) complexes.
02210, United States Cooperativity: A phenomenon in biochemical systems
Bryce Allen − Differentiated Therapeutics, San Diego, where the binding of a molecule to a site on a protein affects
California 92056, United States the affinity of subsequent molecules binding to additional
Woody Sherman − Psivant Therapeutics, Boston, sites on the same protein, leading to either enhanced
Massachusetts 02210, United States; [Link]/0000- (positive cooperativity) or reduced (negative cooperativity)
0001-9079-1376 binding.
DC50: A term used in TPD to denote the concentration
Complete contact information is available at: of a degradation-inducing agent required to achieve 50% of
[Link] its maximum degradation capacity for a specific POI.
Dmax: The maximum degradation capacity of a TPD-
Notes inducing agent representing the highest achievable level of
The authors declare no competing financial interest. POI-degradation under given conditions.
5424 [Link]
J. Chem. Inf. Model. 2023, 63, 5408−5432
Journal of Chemical Information and Modeling [Link]/jcim Perspective

Beyond Rule of 5 (bRo5): Compounds, like (16) Bond, M. J.; Chu, L.; Nalawansha, D. A.; Li, K.; Crews, C. M.
heterobifunctional degraders, that violate Lipinski’s ‘Rule Targeted Degradation of Oncogenic KRAS G12C by VHL-Recruiting
of 5’, which predicts the success of orally absorbed drugs, yet PROTACs. ACS Central Science 2020, 6, 1367−1375.
can be effective therapeutics. (17) Li, X.; et al. c-Myc-Targeting PROTAC Based on a TNA-DNA
Chameleonicity: The ability of a molecule to expose Bivalent Binder for Combination Therapy of Triple-Negative Breast
Cancer. J. Am. Chem. Soc. 2023, 145, 9334−9342.
polar groups in aqueous environments and, through
(18) Nunes, J.; McGonagle, G. A.; Eden, J.; Kiritharan, G.; Touzet,
conformational changes, bury them during membrane M.; Lewell, X.; Emery, J.; Eidam, H.; Harling, J. D.; Anderson, N. A.
permeation thus contributing to its bioavailability and Targeting IRAK4 for Degradation with PROTACs. ACS Med. Chem.
pharmacological profile. Lett. 2019, 10, 1081−1085.
Ternary complex docking: A computational approach to (19) Shiah, J. V.; Grandis, J. R.; Johnson, D. E. Targeting STAT3
study the formation and stability of a complex involving with Proteolysis Targeting Chimeras and Next-Generation Antisense
three molecular entities, such as a POI and an E3 ligase Oligonucleotides. Molecular Cancer Therapeutics 2021, 20, 219−228.
connected by a degrader. (20) Silva, M. C.; Ferguson, F. M.; Cai, Q.; Donovan, K. A.; Nandi,
G.; Patnaik, D.; Zhang, T.; Huang, H.-T.; Lucente, D. E.; Dickerson,

■ REFERENCES
(1) Konstantinova, I. M.; Tsimokha, A. S.; Mittenberg, A. G. Role of
B. C.; Mitchison, T. J.; Fischer, E. S.; Gray, N. S.; Haggarty, S. J.
Targeted degradation of aberrant tau in frontotemporal dementia
patient-derived neuronal cell models. eLife 2019, 8, 45457.
Proteasomes in Cellular Regulation. Int. Rev. Cell Mol. Biol. 2008, 267, (21) Lai, A. C.; Crews, C. M. Induced protein degradation: an
59−124. emerging drug discovery paradigm. Nat. Rev. Drug Discovery 2017, 16,
(2) Myung, J.; Kim, K. B.; Crews, C. M. The ubiquitin-proteasome 101−114.
pathway and proteasome inhibitors. Medicinal research reviews 2001, (22) Liu, Z.; Hu, M.; Yang, Y.; Du, C.; Zhou, H.; Liu, C.; Chen, Y.;
21, 245−273. Fan, L.; Ma, H.; Gong, Y.; Xie, Y. An overview of PROTACs: a
(3) Schneekloth, J. S.; Fonseca, F. N.; Koldobskiy, M.; Mandal, A.; promising drug discovery paradigm. Molecular Biomedicine 2022, 3,
Deshaies, R.; Sakamoto, K.; Crews, C. M. Chemical Genetic Control 46.
of Protein Levels: Selective in Vivo Targeted Degradation. J. Am. (23) Hughes, S.; Ciulli, A. Molecular recognition of ternary
Chem. Soc. 2004, 126, 3748−3754. complexes: a new dimension in the structure-guided design of
(4) Hideshima, T.; Bradner, J. E.; Chauhan, D.; Anderson, K. C. chemical degraders. Essays in Biochemistry 2017, 61, 505−516.
Intracellular Protein Degradation and Its Therapeutic Implications. (24) Chamberlain, P. P.; Hamann, L. G. Development of targeted
Clin. Cancer Res. 2005, 11, 8530−8533. protein degradation therapeutics. Nat. Chem. Biol. 2019, 15, 937−944.
(5) Zhou, P. Targeted protein degradation. Curr. Opin. Chem. Biol. (25) Daurio, N. A.; Zhou, H.; Chen, Y.; Sheth, P. R.; Imbriglio, J. E.;
2005, 9, 51−55. McLaren, D. G.; Tawa, P.; Rachdaoui, N.; Previs, M. J.; Kasumov, T.;
(6) Raina, K.; Crews, C. M. Chemical Inducers of Targeted Protein O’Neil, J.; Previs, S. F. Examining Targeted Protein Degradation from
Degradation*. J. Biol. Chem. 2010, 285, 11057−11060. Physiological and Analytical Perspectives: Enabling Translation
(7) Hershko, A.; Ciechanover, A.; Heller, H.; Haas, A. L.; Rose, I. A. between Cells and Subjects. ACS Chem. Biol. 2020, 15, 2623−2635.
Proposed role of ATP in protein breakdown: conjugation of protein (26) Metzger, M. B.; Pruneda, J. N.; Klevit, R. E.; Weissman, A. M.
with multiple chains of the polypeptide of ATP-dependent RING-type E3 ligases: Master manipulators of E2 ubiquitin-
proteolysis. Proc. Natl. Acad. Sci. U. S. A. 1980, 77, 1783−1786. conjugating enzymes and ubiquitination. Biochimica et Biophysica
(8) Hershko, A.; Heller, H.; Elias, S.; Ciechanover, A. Components Acta (BBA) - Molecular Cell Research 2014, 1843, 47−60.
of ubiquitin-protein ligase system. Resolution, affinity purification, (27) Shabek, N.; Herman-Bachinsky, Y.; Buchsbaum, S.; Lewinson,
and role in protein breakdown. J. Biol. Chem. 1983, 258, 8206−8214. O.; Haj-Yahya, M.; Hejjaoui, M.; Lashuel, H.; Sommer, T.; Brik, A.;
(9) Tamura, T.; Nagy, I.; Lupas, A.; Lottspeich, F.; Cejka, Z.; Ciechanover, A. The Size of the Proteasomal Substrate Determines
Schoofs, G.; Tanaka, K.; De Mot, R.; Baumeister, W. The first Whether Its Degradation Will Be Mediated by Mono- or
characterization of a eubacterial proteasome: the 20S complex of Polyubiquitylation. Mol. Cell 2012, 48, 87−97.
(28) Farnaby, W.; et al. BAF complex vulnerabilities in cancer
Rhodococcus. Curr. Biol. 1995, 5, 766−774.
(10) Lowe, J.; Stock, D.; Jap, B.; Zwickl, P.; Baumeister, W.; Huber, demonstrated via structure-based PROTAC design. Nat. Chem. Biol.
2019, 15, 672−680.
R. Crystal Structure of the 20S Proteasome from the Archaeon T.
(29) Rossi Sebastiano, M.; Doak, B. C.; Backlund, M.;
acidophilum at 3.4 Å Resolution. Science 1995, 268, 533−539.
Poongavanam, V.; Over, B.; Ermondi, G.; Caron, G.; Matsson, P.;
(11) Sakamoto, K. M.; Kim, K. B.; Kumagai, A.; Mercurio, F.;
Kihlberg, J. Impact of Dynamically Exposed Polarity on Permeability
Crews, C. M.; Deshaies, R. J. Protacs: Chimeric molecules that target
and Solubility of Chameleonic Drugs Beyond the Rule of 5. J. Med.
proteins to the Skp1−Cullin−F box complex for ubiquitination and Chem. 2018, 61, 4189−4202.
degradation. Proc. Natl. Acad. Sci. U. S. A. 2001, 98, 8554−8559. (30) Zhou, F.; Chen, L.; Cao, C.; Yu, J.; Luo, X.; Zhou, P.; Zhao, L.;
(12) Gao, X.; et al. Phase 1/2 study of ARV-110, an androgen Du, W.; Cheng, J.; Xie, Y.; Chen, Y. Development of selective mono
receptor (AR) PROTAC degrader, in metastatic castration-resistant or dual PROTAC degrader probe of CDK isoforms. Eur. J. Med.
prostate cancer (mCRPC). Journal of Clinical Oncology 2022, 40, 17− Chem. 2020, 187, 111952.
17. (31) Wang, M.; Lu, J.; Wang, M.; Yang, C.-Y.; Wang, S. Discovery of
(13) Snyder, L. B.; et al. Abstract 44: The discovery of ARV-471, an SHP2-D26 as a First, Potent, and Effective PROTAC Degrader of
orally bioavailable estrogen receptor degrading PROTAC for the SHP2 Protein. J. Med. Chem. 2020, 63, 7510−7528.
treatment of patients with breast cancer. Cancer Res. 2021, 81, 44−44. (32) Tovell, H.; Testa, A.; Zhou, H.; Shpiro, N.; Crafter, C.; Ciulli,
(14) Salami, J.; Alabi, S.; Willard, R. R.; Vitale, N. J.; Wang, J.; Dong, A.; Alessi, D. R. Design and Characterization of SGK3-PROTAC1, an
H.; Jin, M.; McDonnell, D. P.; Crew, A. P.; Neklesa, T. K.; Crews, C. Isoform Specific SGK3 Kinase PROTAC Degrader. ACS Chem. Biol.
M. Androgen receptor degradation by the proteolysis-targeting 2019, 14, 2024−2034.
chimera ARCC-4 outperforms enzalutamide in cellular models of (33) Zhou, H.; et al. Structure-Based Discovery of SD-36 as a
prostate cancer drug resistance. Communications Biology 2018, 1, 100. Potent, Selective, and Efficacious PROTAC Degrader of STAT3
(15) Powell, M.; Blaskovich, M. A. T.; Hansford, K. A. Targeted Protein. J. Med. Chem. 2019, 62, 11280−11300.
Protein Degradation: The New Frontier of Antimicrobial Discovery? (34) Cecchini, C.; Pannilunghi, S.; Tardy, S.; Scapozza, L. From
ACS Infect. Dis. 2021, 7, 2050−2067. Conception to Development: Investigating PROTACs Features for

5425 [Link]
J. Chem. Inf. Model. 2023, 63, 5408−5432
Journal of Chemical Information and Modeling [Link]/jcim Perspective

Improved Cell Permeability and Successful Protein Degradation. (51) Qin, A.-C.; Jin, H.; Song, Y.; Gao, Y.; Chen, Y.-F.; Zhou, L.-N.;
Frontiers in Chemistry 2021, 9, 672267. Wang, S.-S.; Lu, X.-S. The therapeutic effect of the BRD4-degrading
(35) Fisher, S. L.; Phillips, A. J. Targeted protein degradation and PROTAC A1874 in human colon cancer cells. Cell Death & Disease
the enzymology of degraders. Curr. Opin. Chem. Biol. 2018, 44, 47− 2020, 11, 805.
55. (52) Trapotsi, M.-A.; Mouchet, E.; Williams, G.; Monteverde, T.;
(36) Moreau, K.; Coen, M.; Zhang, A. X.; Pachl, F.; Castaldi, M. P.; Juhani, K.; Turkki, R.; Miljković, F.; Martinsson, A.; Mervin, L.;
Dahl, G.; Boyd, H.; Scott, C.; Newham, P. Proteolysis-targeting Pryde, K. R.; Müllers, E.; Barrett, I.; Engkvist, O.; Bender, A.; Moreau,
chimeras in drug development: A safety perspective. Br. J. Pharmacol. K. Cell Morphological Profiling Enables High-Throughput Screening
2020, 177, 1709−1718. for Proteolysis TArgeting Chimera (PROTAC) Phenotypic Signature.
(37) Roy, M. J.; Winkler, S.; Hughes, S. J.; Whitworth, C.; Galant, ACS Chem. Biol. 2022, 17, 1733−1744.
M.; Farnaby, W.; Rumpel, K.; Ciulli, A. SPR-Measured Dissociation (53) Pike, A.; Williamson, B.; Harlfinger, S.; Martin, S.; McGinnity,
Kinetics of PROTAC Ternary Complexes Influence Target D. F. Optimising proteolysis-targeting chimeras (PROTACs) for oral
Degradation Rate. ACS Chem. Biol. 2019, 14, 361−368. drug delivery: a drug metabolism and pharmacokinetics perspective.
(38) Smith, B. E.; Wang, S. L.; Jaime-Figueroa, S.; Harbin, A.; Wang, Drug Discovery Today 2020, 25, 1793−1800.
J.; Hamman, B. D.; Crews, C. M. Differential PROTAC substrate (54) Goracci, L.; Desantis, J.; Valeri, A.; Castellani, B.; Eleuteri, M.;
specificity dictated by orientation of recruited E3 ligase. Nat. Cruciani, G. Understanding the Metabolism of Proteolysis Targeting
Commun. 2019, 10, 131. Chimeras (PROTACs): The Next Step toward Pharmaceutical
(39) Zorba, A.; Nguyen, C.; Xu, Y.; Starr, J.; Borzilleri, K.; Smith, J.; Applications. J. Med. Chem. 2020, 63, 11615−11638.
Zhu, H.; Farley, K. A.; Ding, W.; Schiemer, J.; et al. Delineating the (55) Cantrill, C.; Chaturvedi, P.; Rynn, C.; Petrig Schaffland, J.;
role of cooperativity in the design of potent PROTACs for BTK. Proc. Walter, I.; Wittwer, M. B. Fundamental aspects of DMPK
Natl. Acad. Sci. U. S. A. 2018, 115, E7285−E7292. optimization of targeted protein degraders. Drug discovery today
(40) Gadd, M. S.; Testa, A.; Lucas, X.; Chan, K.-H.; Chen, W.; 2020, 25, 969−982.
Lamont, D. J.; Zengerle, M.; Ciulli, A. Structural basis of PROTAC (56) Di, L.; Fish, P. V.; Mano, T. Bridging solubility between drug
cooperative recognition for selective protein degradation. Nat. Chem. discovery and development. Drug discovery today 2012, 17, 486−495.
Biol. 2017, 13, 514−521. (57) Dahan, A.; Miller, J. M. The Solubility-Permeability Interplay
(41) Zoppi, V.; Hughes, S. J.; Maniaci, C.; Testa, A.; Gmaschitz, T.; and Its Implications in Formulation Design and Development for
Wieshofer, C.; Koegl, M.; Riching, K. M.; Daniels, D. L.; Spallarossa, Poorly Soluble Drugs. AAPS Journal 2012, 14, 244−251.
A.; Ciulli, A. Iterative Design and Optimization of Initially Inactive (58) Naylor, M. R.; et al. Lipophilic Permeability Efficiency
Proteolysis Targeting Chimeras (PROTACs) Identify VZ185 as a Reconciles the Opposing Roles of Lipophilicity in Membrane
Potent, Fast, and Selective von Hippel-Lindau (VHL) Based Dual
Permeability and Aqueous Solubility. J. Med. Chem. 2018, 61,
Degrader Probe of BRD9 and BRD7. J. Med. Chem. 2019, 62, 699−
11169−11182.
726.
(59) Whitty, A.; Zhong, M.; Viarengo, L.; Beglov, D.; Hall, D. R.;
(42) Lu, M.; Liu, T.; Jiao, Q.; Ji, J.; Tao, M.; Liu, Y.; You, Q.; Jiang,
Vajda, S. Quantifying the chameleonic properties of macrocycles and
Z. Discovery of a Keap1-dependent peptide PROTAC to knockdown
other high-molecular-weight drugs. Drug Discovery Today 2016, 21,
Tau by ubiquitination-proteasome degradation pathway. European
712−717.
journal of medicinal chemistry 2018, 146, 251−259.
(60) Danelius, E.; Poongavanam, V.; Peintner, S.; Wieske, L. H. E.;
(43) Testa, A.; Hughes, S. J.; Lucas, X.; Wright, J. E.; Ciulli, A.
Erdélyi, M.; Kihlberg, J. Solution Conformations Explain the
Structure-Based Design of a Macrocyclic PROTAC. Angew. Chem.,
Int. Ed. 2020, 59, 1727−1734. Chameleonic Behaviour of Macrocyclic Drugs. Chem.�Eur. J. 2020,
(44) Bai, L.; Zhou, H.; Xu, R.; Zhao, Y.; Chinnaswamy, K.; 26, 5231−5244.
McEachern, D.; Chen, J.; Yang, C.-Y.; Liu, Z.; Wang, M.; et al. A (61) Lipinski, C. A.; Lombardo, F.; Dominy, B. W.; Feeney, P. J.
Potent and Selective Small-Molecule Degrader of STAT3 Achieves Experimental and computational approaches to estimate solubility and
Complete Tumor Regression In Vivo. Cancer Cell 2019, 36, 498− permeability in drug discovery and development settings1PII of
511.e17. original article: S0169−409X(96)00423−1. The article was originally
(45) Nowak, R. P.; DeAngelo, S. L.; Buckley, D.; He, Z.; Donovan, published in Advanced Drug Delivery Reviews 23 (1997) 3−25.1.
K. A.; An, J.; Safaee, N.; Jedrychowski, M. P.; Ponthier, C. M.; Ishoey, Adv. Drug Delivery Rev. 2001, 46, 3−26.
M.; Zhang, T.; Mancias, J. D.; Gray, N. S.; Bradner, J. E.; Fischer, E. S. (62) Veber, D. F.; Johnson, S. R.; Cheng, H.-Y.; Smith, B. R.; Ward,
Plasticity in binding confers selectivity in ligand-induced protein K. W.; Kopple, K. D. Molecular Properties That Influence the Oral
degradation. Nat. Chem. Biol. 2018, 14, 706−714. Bioavailability of Drug Candidates. J. Med. Chem. 2002, 45, 2615−
(46) Lu, J.; Qian, Y.; Altieri, M.; Dong, H.; Wang, J.; Raina, K.; 2623.
Hines, J.; Winkler, J. D.; Crew, A. P.; Coleman, K.; Crews, C. M. (63) Matsson, P.; Kihlberg, J. How Big Is Too Big for Cell
Hijacking the E3 Ubiquitin Ligase Cereblon to Efficiently Target Permeability? J. Med. Chem. 2017, 60, 1662−1664.
BRD4. Chemistry & Biology 2015, 22, 755−763. (64) Poongavanam, V.; Doak, B. C.; Kihlberg, J. Opportunities and
(47) Lai, A. C.; Toure, M.; Hellerschmied, D.; Salami, J.; Jaime- guidelines for discovery of orally absorbed drugs in beyond rule of 5
Figueroa, S.; Ko, E.; Hines, J.; Crews, C. M. Modular PROTAC space. Curr. Opin. Chem. Biol. 2018, 44, 23−29.
Design for the Degradation of Oncogenic BCR-ABL. Angew. Chem., (65) Ermondi, G.; Vallaro, M.; Goetz, G.; Shalaeva, M.; Caron, G.
Int. Ed. 2016, 55, 807−810. Updating the portfolio of physicochemical descriptors related to
(48) Riching, K. M.; Mahan, S.; Corona, C. R.; McDougall, M.; permeability in the beyond the rule of 5 chemical space. European
Vasta, J. D.; Robers, M. B.; Urh, M.; Daniels, D. L. Quantitative Live- Journal of Pharmaceutical Sciences 2020, 146, 105274.
Cell Kinetic Degradation and Mechanistic Profiling of PROTAC (66) Poongavanam, V.; Kihlberg, J. PROTAC cell permeability and
Mode of Action. ACS Chem. Biol. 2018, 13, 2758−2770. oral bioavailability: a journey into uncharted territory. Future
(49) Webb, T.; Craigon, C.; Ciulli, A. Targeting epigenetic Medicinal Chemistry 2022, 14 (3), 123−126.
modulators using PROTAC degraders: Current status and future (67) Avdeef, A.; Testa, B. Physicochemical profiling in drug
perspective. Bioorg. Med. Chem. Lett. 2022, 63, 128653. research: a brief survey of the state-of-the-art of experimental
(50) Burslem, G. M.; Smith, B. E.; Lai, A. C.; Jaime-Figueroa, S.; techniques. Cellular and Molecular Life Sciences CMLS 2002, 59,
McQuaid, D. C.; Bondeson, D. P.; Toure, M.; Dong, H.; Qian, Y.; 1681−1689.
Wang, J.; Crew, A. P.; Hines, J.; Crews, C. M. The Advantages of (68) Kerns, E. H.; Di, L. Automation in Pharmaceutical Profiling.
Targeted Protein Degradation Over Inhibition: An RTK Case Study. JALA: Journal of the Association for Laboratory Automation 2005, 10,
Cell Chemical Biology 2018, 25, 67−77.e3. 114−123.

5426 [Link]
J. Chem. Inf. Model. 2023, 63, 5408−5432
Journal of Chemical Information and Modeling [Link]/jcim Perspective

(69) Bunally, S. B.; Luscombe, C. N.; Young, R. J. Using Fluorescence Resonance Energy-Transfer Assay. ACS Omega 2021,
Physicochemical Measurements to Influence Better Compound 6, 680−695.
Design. SLAS Discovery 2019, 24, 791−801. (87) Beaudet, L.; Rodriguez-Suarez, R.; Venne, M.-H.; Caron, M.;
(70) Lombardo, F.; Shalaeva, M. Y.; Tupper, K. A.; Gao, F.; Bédard, J.; Brechler, V.; Parent, S.; Bielefeld-Sévigny, M. AlphaLISA
Abraham, M. H. ElogPoct: ATool for Lipophilicity Determination in immunoassays: the no-wash alternative to ELISAs for research and
Drug Discovery. J. Med. Chem. 2000, 43, 2922−2928. drug discovery. Nat. Methods 2008, 5, an8−an9.
(71) Lombardo, F.; Shalaeva, M. Y.; Tupper, K. A.; Gao, F. (88) Fredriksson, S.; Gullberg, M.; Jarvius, J.; Olsson, C.; Pietras, K.;
ElogDoct: ATool for Lipophilicity Determination in Drug Discovery. Gústafsdóttir, S. M.; Ö stman, A.; Landegren, U. Protein detection
2. Basic and Neutral Compounds. J. Med. Chem. 2001, 44, 2490− using proximity-dependent DNA ligation assays. Nat. Biotechnol.
2497. 2002, 20, 473−477.
(72) Garcia Jimenez, D.; Rossi Sebastiano, M.; Vallaro, M.; Mileo, (89) Bondeson, D. P.; Smith, B. E.; Burslem, G. M.; Buhimschi, A.
V.; Pizzirani, D.; Moretti, E.; Ermondi, G.; Caron, G. Designing D.; Hines, J.; Jaime-Figueroa, S.; Wang, J.; Hamman, B. D.;
Soluble PROTACs: Strategies and Preliminary Guidelines. J. Med. Ishchenko, A.; Crews, C. M. Lessons in PROTAC Design from
Chem. 2022, 65 (19), 12639−12649. Selective Degradation with a Promiscuous Warhead. Cell Chemical
(73) Pöstges, F.; Kayser, K.; Appelhaus, J.; Monschke, M.; Biology 2018, 25 (1), 78−87.e5.
Gütschow, M.; Steinebach, C.; Wagner, K. G. Solubility Enhanced (90) Bortoluzzi, A.; Ciulli, A. Applied Biophysics for Drug Discovery
Formulation Approaches to Overcome Oral Delivery Obstacles of 2017, 153−173.
PROTACs. Pharmaceutics 2023, 15 (1), 156. (91) Harner, M. J.; Frank, A. O.; Fesik, S. W. Fragment-based drug
(74) Avdeef, A. Absorption and Drug Development: Solubility, discovery using NMR spectroscopy. Journal of Biomolecular NMR
Permeability, and Charge State; John Wiley & Sons, Ltd, 2012; 2013, 56, 65−75.
Chapter 6, pp 251−318. (92) Dias, D. M.; Van Molle, I.; Baud, M. G. J.; Galdeano, C.;
(75) Avdeef, A.; Fuguet, E.; Llinàs, A.; Ràfols, C.; Bosch, E.; Völgyi, Geraldes, C. F. G. C.; Ciulli, A. Is NMR Fragment Screening Fine-
G.; Verbić, T.; Boldyreva, E.; Takács-Novák, K. Equilibrium solubility Tuned to Assess Druggability of Protein-Protein Interactions? ACS
measurement of ionizable drugs - consensus recommendations for Med. Chem. Lett. 2014, 5, 23−28.
improving data quality. ADMET and DMPK 2016, 4, 117−178. (93) Lucas, X.; Van Molle, I.; Ciulli, A. Surface Probing by
(76) Lin, X.; Hu, Y.; Liu, L.; Su, L.; Li, N.; Yu, J.; Tang, B.; Yang, Z. Fragment-Based Screening and Computational Methods Identifies
Physical Stability of Amorphous Solid Dispersions: a Physicochemical Ligandable Pockets on the von Hippel-Lindau (VHL) E3 Ubiquitin
Perspective with Thermodynamic, Kinetic and Environmental Ligase. Journal of medicinal chemistry 2018, 61, 7387−7393.
Aspects. Pharm. Res. 2018, 35 (6), 125. (94) Buckley, D. L.; Van Molle, I.; Gareiss, P. C.; Tae, H. S.; Michel,
(77) Kansy, M.; Avdeef, A.; Fischer, H. Advances in screening for J.; Noblin, D. J.; Jorgensen, W. L.; Ciulli, A.; Crews, C. M. Targeting
membrane permeability: high-resolution PAMPA for medicinal the von Hippel-Lindau E3 Ubiquitin Ligase Using Small Molecules
chemists. Drug Discovery Today: Technologies 2004, 1, 349−355. To Disrupt the VHL/HIF-1α Interaction. J. Am. Chem. Soc. 2012,
(78) Berben, P.; Bauer-Brandl, A.; Brandl, M.; Faller, B.; Flaten, G. 134, 4465−4468.
E.; Jacobsen, A.-C.; Brouwers, J.; Augustijns, P. Drug permeability (95) de Castro, G. V.; Ciulli, A. Estimating the cooperativity of
profiling using cell-free permeation tools: Overview and applications. PROTAC-induced ternary complexes using 19F NMR displacement
European Journal of Pharmaceutical Sciences 2018, 119, 219−233. assay. RSC Med. Chem. 2021, 12, 1765−1770.
(79) Hubatsch, I.; Ragnarsson, E. G. E.; Artursson, P. Determination (96) Wijaya, A. J.; Farnaby, W.; Ciulli, A. Crystallization of VHL-
of drug permeability and prediction of drug absorption in Caco-2 based PROTAC-induced ternary complexes. Methods Enzymol. 2023,
monolayers. Nat. Protoc. 2007, 2, 2111−2119. 681, 241−263.
(80) Rathod, D.; Fu, Y.; Patel, K. BRD4 PROTAC as a novel (97) Kung, W.-W.; Ramachandran, S.; Makukhin, N.; Bruno, E.;
therapeutic approach for the treatment of vemurafenib resistant Ciulli, A. Structural insights into substrate recognition by the SOCS2
melanoma: Preformulation studies, formulation development and in E3 ubiquitin ligase. Nat. Commun. 2019, 10, 2534.
vitro evaluation. European journal of pharmaceutical sciences: official (98) Bussiere, D. E.; et al. Structural basis of indisulam-mediated
journal of the European Federation for Pharmaceutical Sciences 2019, RBM39 recruitment to DCAF15 E3 ligase complex. Nat. Chem. Biol.
138, 105039. 2020, 16, 15−23.
(81) Klein, V. G.; Townsend, C. E.; Testa, A.; Zengerle, M.; (99) Renaud, J.-P.; Chari, A.; Ciferri, C.; Liu, W.-t.; Rémigy, H.-W.;
Maniaci, C.; Hughes, S. J.; Chan, K.-H.; Ciulli, A.; Lokey, R. S. Stark, H.; Wiesmann, C. Cryo-EM in drug discovery: achievements,
Understanding and Improving the Membrane Permeability of limitations and prospects. Nat. Rev. Drug Discovery 2018, 17, 471−
VH032-Based PROTACs. ACS Med. Chem. Lett. 2020, 11, 1732− 492.
1738. (100) Maniaci, C.; Hughes, S. J.; Testa, A.; Chen, W.; Lamont, D. J.;
(82) Foley, C. A.; Potjewyd, F.; Lamb, K. N.; James, L. I.; Frye, S. V. Rocha, S.; Alessi, D. R.; Romeo, R.; Ciulli, A. Homo-PROTACs:
Assessing the Cell Permeability of Bivalent Chemical Degraders Using bivalent small-molecule dimerizers of the VHL E3 ubiquitin ligase to
the Chloroalkane Penetration Assay. ACS Chem. Biol. 2020, 15, 290− induce self-degradation. Nat. Commun. 2017, 8, 830.
295. (101) Beveridge, R.; Kessler, D.; Rumpel, K.; Ettmayer, P.;
(83) Zeng, M.; Xiong, Y.; Safaee, N.; Nowak, R.ła. P.; Donovan, K. Meinhart, A.; Clausen, T. Native Mass Spectrometry Can Effectively
A.; Yuan, C. J.; Nabet, B.; Gero, T. W.; Feru, F.; Li, L.; et al. Exploring Predict PROTAC Efficacy. ACS Central Science 2020, 6, 1223−1230.
Targeted Degradation Strategy for Oncogenic KRASG12C. Cell (102) Jafari, R.; Almqvist, H.; Axelsson, H.; Ignatushchenko, M.;
Chemical Biology 2020, 27 (1), 19−31.e6. Lundbäck, T.; Nordlund, P.; Molina, D. M. The cellular thermal shift
(84) Atilaw, Y.; Poongavanam, V.; Svensson Nilsson, C.; Nguyen, assay for evaluating drug target interactions in cells. Nat. Protoc. 2014,
D.; Giese, A.; Meibom, D.; Erdelyi, M.; Kihlberg, J. Solution 9, 2100−2122.
Conformations Shed Light on PROTAC Cell Permeability. ACS Med. (103) Yeung, W.; Luo, G.; Wang, Q.; Ou, J. Capillary electro-
Chem. Lett. 2021, 12, 107−114. phoresis-based immunoassay. Journal of Chromatography B 2003, 797,
(85) Rossi, A. M.; Taylor, C. W. Analysis of protein-ligand 217−228.
interactions by fluorescence polarization. Nat. Protoc. 2011, 6, 365− (104) Li, K. S.; Schaper Bergman, E. T.; Beno, B. R.; Huang, R. Y.-
387. C.; Deyanova, E.; Chen, G.; Gross, M. L. Hydrogen-Deuterium
(86) Lin, W.; Li, Y.; Yang, L.; Chen, T. Development of BODIPY FL Exchange and Hydroxyl Radical Footprinting for Mapping Hydro-
VH032 as a High-Affinity and Selective von Hippel-Lindau E3 Ligase phobic Interactions of Human Bromodomain with a Small Molecule
Fluorescent Probe and Its Application in a Time-Resolved Inhibitor. J. Am. Soc. Mass Spectrom. 2019, 30, 2795−2804.

5427 [Link]
J. Chem. Inf. Model. 2023, 63, 5408−5432
Journal of Chemical Information and Modeling [Link]/jcim Perspective

(105) Chalmers, M. J.; Busby, S. A.; Pascal, B. D.; He, Y.; (123) Verma, J.; Khedkar, V.; Coutinho, E. 3D-QSAR in Drug
Hendrickson, C. L.; Marshall, A. G.; Griffin, P. R. Probing Protein Design - A Review. Current Topics in Medicinal Chemistry 2010, 10,
Ligand Interactions by Automated Hydrogen/Deuterium Exchange 95−115.
Mass Spectrometry. Anal. Chem. 2006, 78, 1005−1014. (124) Kearnes, S.; McCloskey, K.; Berndl, M.; Pande, V.; Riley, P.
(106) Johnson, D. T.; Di Stefano, L. H.; Jones, L. M. Fast Molecular graph convolutions: moving beyond fingerprints. Journal of
photochemical oxidation of proteins (FPOP): A powerful mass Computer-Aided Molecular Design 2016, 30, 595−608.
spectrometry-based structural proteomics tool. J. Biol. Chem. 2019, (125) Feinberg, E. N.; Sur, D.; Wu, Z.; Husic, B. E.; Mai, H.; Li, Y.;
294, 11969−11979. Sun, S.; Yang, J.; Ramsundar, B.; Pande, V. S. PotentialNet for
(107) Eron, S. J.; Huang, H.; Agafonov, R. V.; Fitzgerald, M. E.; Molecular Property Prediction. ACS Central Science 2018, 4, 1520−
Patel, J.; Michael, R. E.; Lee, T. D.; Hart, A. A.; Shaulsky, J.; 1530.
Nasveschuk, C. G.; Phillips, A. J.; Fisher, S. L.; Good, A. Structural (126) Yang, K.; Swanson, K.; Jin, W.; Coley, C.; Eiden, P.; Gao, H.;
Characterization of Degrader-Induced Ternary Complexes Using Guzman-Perez, A.; Hopper, T.; Kelley, B.; Mathea, M.; Palmer, A.;
Hydrogen-Deuterium Exchange Mass Spectrometry and Computa- Settels, V.; Jaakkola, T.; Jensen, K.; Barzilay, R. Analyzing Learned
tional Modeling: Implications for Structure-Based Design. ACS Chem. Molecular Representations for Property Prediction. J. Chem. Inf.
Biol. 2021, 16, 2228−2243. Model. 2019, 59, 3370−3388.
(108) Dixon, T.; MacPherson, D.; Mostofian, B.; Dauzhenka, T.; (127) Bergström, C. A.; Charman, W. N.; Porter, C. J. Computa-
Lotz, S.; McGee, D.; Shechter, S.; Shrestha, U. R.; Wiewiora, R.; tional prediction of formulation strategies for beyond-rule-of-5
McDargh, Z. A.; et al. Predicting the structural basis of targeted compounds. Adv. Drug Delivery Rev. 2016, 101, 6−21.
protein degradation by integrating molecular dynamics simulations (128) Maple, H. J.; Clayden, N.; Baron, A.; Stacey, C.; Felix, R.
with structural mass spectrometry. Nat. Commun. 2022, 13, 5884. Developing degraders: principles and perspectives on design and
(109) Grandi, P.; Bantscheff, M. Advanced proteomics approaches chemical space. MedChemComm 2019, 10, 1755−1764.
to unravel protein homeostasis. Drug discovery today. Technologies (129) Edmondson, S. D.; Yang, B.; Fallan, C. Proteolysis Targeting
2019, 31, 99−108. Chimeras (PROTACs) in ‘Beyond Rule-of-Five’ Chemical Space:
(110) Steger, M.; Karayel, Ö .; Demichev, V. Ubiquitinomics: Recent Progress and Future Challenges. Bioorg. Med. Chem. Lett.
History, methods, and applications in basic research and drug 2019, 29, 1555−1564.
discovery. PROTEOMICS 2022, 22, 2200074. (130) Jimenez, D. G.; Sebastiano, M. R.; Caron, G.; Ermondi, G. Are
(111) Khan, S.; et al. A selective BCL-XL PROTAC degrader we ready to design oral PROTACs®? ADMET and DMPK 2021, 9,
achieves safe and potent antitumor activity. Nature Medicine 2019, 25, 243−254.
1938−1947. (131) Evans, D. J.; Yovanno, R. A.; Rahman, S.; Cao, D. W.; Beckett,
(112) Ottis, P.; Toure, M.; Cromm, P. M.; Ko, E.; Gustafson, J. L.; M. Q.; Patel, M. H.; Bandak, A. F.; Lau, A. Y. Finding Druggable Sites
Crews, C. M. Assessing Different E3 Ligases for Small Molecule
in Proteins Using TACTICS. J. Chem. Inf. Model. 2021, 61, 2897−
Induced Protein Ubiquitination and Degradation. ACS Chem. Biol.
2910.
2017, 12, 2570−2578.
(132) Tze-Yang Ng, J.; Tan, Y. S. Accelerated Ligand-Mapping
(113) Savitski, M. M.; Zinn, N.; Faelth-Savitski, M.; Poeckel, D.;
Molecular Dynamics Simulations for the Detection of Recalcitrant
Gade, S.; Becher, I.; Muelbaier, M.; Wagner, A. J.; Strohmer, K.;
Cryptic Pockets and Occluded Binding Sites. J. Chem. Theory Comput.
Werner, T.; et al. Multiplexed Proteome Dynamics Profiling Reveals
2022, 18, 1969−1981.
Mechanisms Controlling Protein Homeostasis. Cell 2018, 173, 260−
(133) Meller, A.; Ward, M.; Borowsky, J.; Kshirsagar, M.;
274.e25.
(114) Zhang, A. X.; Cassidy, K.; Dahl, G.; Moreau, K.; Pachl, F.; Lotthammer, J. M.; Oviedo, F.; Ferres, J. L.; Bowman, G. R.
Zuhl, A. M. The Vital Role of Proteomics in Characterizing Novel Predicting locations of cryptic pockets from single protein structures
Protein Degraders. SLAS DISCOVERY: Advancing the Science of Drug using the PocketMiner graph neural network. Nat. Commun. 2023, 14,
Discovery 2021, 26, 518−523. 1177.
(115) Riching, K. M.; Schwinn, M. K.; Vasta, J. D.; Robers, M. B.; (134) Amaro, R. E.; Baudry, J.; Chodera, J.; Demir; McCammon, J.
Machleidt, T.; Urh, M.; Daniels, D. L. CDK Family PROTAC A.; Miao, Y.; Smith, J. C. Ensemble Docking in Drug Discovery.
Profiling Reveals Distinct Kinetic Responses and Cell Cycle- Biophys. J. 2018, 114, 2271−2278.
Dependent Degradation of CDK2. SLAS Discovery: Advancing the (135) Salmaso, V.; Moro, S. Bridging Molecular Docking to
Science of Drug Discovery 2021, 26, 560−569. Molecular Dynamics in Exploring Ligand-Protein Recognition
(116) Liu, X.; Zhang, X.; Lv, D.; Yuan, Y.; Zheng, G.; Zhou, D. Process: An Overview. Frontiers in Pharmacology 2018, 9, 923.
Assays and technologies for developing proteolysis targeting chimera (136) Śledź, P.; Caflisch, A. Protein structure-based drug design:
degraders. Future Medicinal Chemistry 2020, 12, 1155−1179. from docking to molecular dynamics. Curr. Opin. Struct. Biol. 2018,
(117) Ward, J. A.; Perez-Lopez, C.; Mayor-Ruiz, C. Biophysical and 48, 93−102.
computational approaches to study ternary complexes: A ‘cooperative (137) Cournia, Z.; Allen, B.; Sherman, W. Relative Binding Free
relationship’ to rationalize targeted protein degradation. ChemBio- Energy Calculations in Drug Discovery: Recent Advances and
Chem. 2023, 24, e202300163. Practical Considerations. J. Chem. Inf. Model. 2017, 57, 2911−2937.
(118) Cherkasov, A.; et al. QSAR Modeling: Where Have You Been? (138) Cournia, Z.; Allen, B. K.; Beuming, T.; Pearlman, D. A.;
Where Are You Going To? J. Med. Chem. 2014, 57, 4977−5010. Radak, B. K.; Sherman, W. Rigorous Free Energy Simulations in
(119) Shen, M.; Xiao, Y.; Golbraikh, A.; Gombar, V. K.; Tropsha, A. Virtual Screening. J. Chem. Inf. Model. 2020, 60, 4153−4169.
Development and Validation of k-Nearest-Neighbor QSPR Models of (139) Xu, H. The slow but steady rise of binding free energy
Metabolic Stability of Drug Candidates. J. Med. Chem. 2003, 46, calculations in drug discovery. Journal of Computer-Aided Molecular
3013−3020. Design 2023, 37, 67−74.
(120) Cano, G.; Garcia-Rodriguez, J.; Garcia-Garcia, A.; Perez- (140) Decherchi, S.; Cavalli, A. Thermodynamics and Kinetics of
Sanchez, H.; Benediktsson, J. A.; Thapa, A.; Barr, A. Automatic Drug-Target Binding by Molecular Simulation. Chem. Rev. 2020, 120,
selection of molecular descriptors using random forest: Application to 12788−12833.
drug discovery. Expert Systems with Applications 2017, 72, 151−159. (141) Bergström, C. A. S.; Norinder, U.; Luthman, K.; Artursson, P.
(121) Heikamp, K.; Bajorath, J. Support vector machines for drug Molecular Descriptors Influencing Melting Point and Their Role in
discovery. Expert Opinion on Drug Discovery 2014, 9, 93−104. Classification of Solid Drugs. J. Chem. Inf. Comput. Sci. 2003, 43,
(122) Zhang, L.; Tan, J.; Han, D.; Zhu, H. From machine learning to 1177−1185.
deep learning: progress in machine intelligence for rational drug (142) Leo, A.; Hansch, C.; Elkins, D. Partition coefficients and their
discovery. Drug Discovery Today 2017, 22, 1680−1685. uses. Chem. Rev. 1971, 71, 525−616.

5428 [Link]
J. Chem. Inf. Model. 2023, 63, 5408−5432
Journal of Chemical Information and Modeling [Link]/jcim Perspective

(143) McDonagh, J. L.; Nath, N.; De Ferrari, L.; van Mourik, T.; (161) Rafi, S. B.; Hearn, B. R.; Vedantham, P.; Jacobson, M. P.;
Mitchell, J. B. O. Uniting Cheminformatics and Chemical Theory To Renslo, A. R. Predicting and Improving the Membrane Permeability
Predict the Intrinsic Aqueous Solubility of Crystalline Druglike of Peptidic Small Molecules. J. Med. Chem. 2012, 55, 3163−3169.
Molecules. J. Chem. Inf. Model. 2014, 54, 844−856. (162) Witek, J.; Wang, S.; Schroeder, B.; Lingwood, R.; Dounas, A.;
(144) Raevsky, O. A.; Grigorev, V. Y.; Polianczyk, D. E.; Raevskaja, Roth, H.-J.; Fouché, M.; Blatter, M.; Lemke, O.; Keller, B.; Riniker, S.
O. E.; Dearden, J. C. Aqueous Drug Solubility: What Do We Measure, Rationalization of the Membrane Permeability Differences in a Series
Calculate and QSPR Predict? Mini-Reviews in Medicinal Chemistry of Analogue Cyclic Decapeptides. J. Chem. Inf. Model. 2019, 59, 294−
2019, 19, 362−372. 308.
(145) Michielan, L.; Bacilieri, M.; Kaseda, C.; Moro, S. Prediction of (163) Wang, C. K.; Swedberg, J. E.; Harvey, P. J.; Kaas, Q.; Craik, D.
the aqueous solvation free energy of organic compounds by using J. Conformational Flexibility Is a Determinant of Permeability for
autocorrelation of molecular electrostatic potential surface properties Cyclosporin. J. Phys. Chem. B 2018, 122, 2261−2276.
combined with response surface analysis. Bioorganic & medicinal (164) Over, B.; et al. Structural and conformational determinants of
chemistry 2008, 16, 5733−42. macrocycle cell permeability. Nat. Chem. Biol. 2016, 12, 1065−1074.
(146) Tjong, H.; Zhou, H.-X. Prediction of Protein Solubility from (165) Poongavanam, V.; Atilaw, Y.; Ye, S.; Wieske, L. H.; Erdelyi,
Calculation of Transfer Free Energy. Biophys. J. 2008, 95, 2601−2609. M.; Ermondi, G.; Caron, G.; Kihlberg, J. Predicting the Permeability
(147) Palmer, D. S.; Frolov, A. I.; Ratkova, E. L.; Fedorov, M. V. of Macrocycles from Conformational Sampling - Limitations of
Toward a Universal Model To Calculate the Solvation Thermody- Molecular Flexibility. J. Pharm. Sci. 2021, 110, 301−313.
namics of Druglike Molecules: The Importance of New Experimental (166) Poongavanam, V.; Atilaw, Y.; Siegel, S.; Giese, A.; Lehmann,
Databases. Mol. Pharmaceutics 2011, 8, 1423−1429. L.; Meibom, D.; Erdelyi, M.; Kihlberg, J. Linker-Dependent Folding
(148) Belmares, M.; Blanco, M.; Goddard, W. A.; Ross, R. B.; Rationalizes PROTAC Cell Permeability. J. Med. Chem. 2022, 65,
Caldwell, G.; Chou, S.; Pham, J.; Olofson, P. M.; Thomas, C. 13029−13040.
Hildebrand and Hansen solubility parameters from Molecular (167) Swift, R. V.; Amaro, R. E. Back to the Future: Can Physical
Dynamics with applications to electronic nose polymer sensors. J. Models of Passive Membrane Permeability Help Reduce Drug
Comput. Chem. 2004, 25, 1814−1826. Candidate Attrition and Move Us Beyond QSPR? Chemical Biology
(149) Paluch, A. S.; Parameswaran, S.; Liu, S.; Kolavennu, A.; & Drug Design 2013, 81, 61−71.
Mobley, D. L. Predicting the excess solubility of acetanilide, (168) Awoonor-Williams, E.; Rowley, C. N. Molecular simulation of
acetaminophen, phenacetin, benzocaine, and caffeine in binary nonfacilitated membrane permeation. Biochimica et Biophysica Acta
water/ethanol mixtures via molecular simulation. J. Chem. Phys. (BBA) - Biomembranes 2016, 1858, 1672−1687.
(169) Venable, R. M.; Krämer, A.; Pastor, R. W. Molecular
2015, 142, 044508.
Dynamics Simulations of Membrane Permeability. Chem. Rev. 2019,
(150) Hossain, S.; Kabedev, A.; Parrow, A.; Bergström, C.; Larsson,
119, 5954−5997.
P. Molecular simulation as a computational pharmaceutics tool to
(170) Lee, C. T.; Comer, J.; Herndon, C.; Leung, N.; Pavlova, A.;
predict drug solubility, solubilization processes and partitioning. Eur.
Swift, R. V.; Tung, C.; Rowley, C. N.; Amaro, R. E.; Chipot, C.;
J. Pharm. Biopharm. 2019, 137, 46−55.
Wang, Y.; Gumbart, J. C. Simulation-Based Approaches for
(151) Boobier, S.; Hose, D. R. J.; Blacker, A. J.; Nguyen, B. N.
Determining Membrane Permeability of Small Compounds. J.
Machine learning with physicochemical relationships: solubility
Chem. Inf. Model. 2016, 56, 721−733.
prediction in organic solvents and water. Nat. Commun. 2020, 11, (171) Votapka, L. W.; Lee, C. T.; Amaro, R. E. Two Relations to
5753. Estimate Membrane Permeability Using Milestoning. J. Phys. Chem. B
(152) Ye, Z.; Ouyang, D. Prediction of small-molecule compound 2016, 120, 8606−8616.
solubility in organic solvents by machine learning algorithms. Journal (172) Bennion, B. J.; Be, N. A.; McNerney, M. W.; Lao, V.; Carlson,
of Cheminformatics 2021, 13, 98. E. M.; Valdez, C. A.; Malfatti, M. A.; Enright, H. A.; Nguyen, T. H.;
(153) Wieder, O.; Kuenemann, M.; Wieder, M.; Seidel, T.; Meyer, Lightstone, F. C.; Carpenter, T. S. Predicting a Drug’s Membrane
C.; Bryant, S. D.; Langer, T. Improved Lipophilicity and Aqueous Permeability: A Computational Model Validated With in Vitro
Solubility Prediction with Composite Graph. Molecules 2021, 26, Permeability Assay Data. J. Phys. Chem. B 2017, 121, 5228−5237.
6185. (173) Sugita, M.; Sugiyama, S.; Fujie, T.; Yoshikawa, Y.; Yanagisawa,
(154) Ermondi, G.; Poongavanam, V.; Vallaro, M.; Kihlberg, J.; K.; Ohue, M.; Akiyama, Y. Large-Scale Membrane Permeability
Caron, G. Solubility prediction in the bRo5 chemical space: where are Prediction of Cyclic Peptides Crossing a Lipid Bilayer Based on
we right now? ADMET and DMPK 2020, 0, 207−214. Enhanced Sampling Molecular Dynamics Simulations. J. Chem. Inf.
(155) Ermondi, G.; Vallaro, M.; Caron, G. Degraders early Model. 2021, 61, 3681−3695.
developability assessment: face-to-face with molecular properties. (174) Wadhwa, R.; Yadav, N. S.; Katiyar, S. P.; Yaguchi, T.; Lee, C.;
Drug Discovery Today 2020, 25, 1585−1591. Ahn, H.; Yun, C.-O.; Kaul, S. C.; Sundar, D. Molecular dynamics
(156) DeGoey, D. A.; Chen, H.-J.; Cox, P. B.; Wendt, M. D. Beyond simulations and experimental studies reveal differential permeability
the Rule of 5: Lessons Learned from AbbVie’s Drugs and Compound of withaferin-A and withanone across the model cell membrane. Sci.
Collection. J. Med. Chem. 2018, 61, 2636−2651. Rep. 2021, 11, 2352.
(157) Avdeef, A.; Kansy, M. Flexible-Acceptor” General Solubility (175) Zhang, S.; Thompson, J. P.; Xia, J.; Bogetti, A. T.; York, F.;
Equation for beyond Rule of 5 Drugs. Mol. Pharmaceutics 2020, 17, Skillman, A. G.; Chong, L. T.; LeBard, D. N. Mechanistic Insights into
3930−3940. Passive Membrane Permeability of Drug-like Molecules from a
(158) Rezai, T.; Yu, B.; Millhauser, G. L.; Jacobson, M. P.; Lokey, R. Weighted Ensemble of Trajectories. J. Chem. Inf. Model. 2022, 62,
S. Testing the Conformational Hypothesis of Passive Membrane 1891−1904.
Permeability Using Synthetic Cyclic Peptide Diastereomers. J. Am. (176) Marrink, S. J.; Corradi, V.; Souza, P. C.; Ingólfsson, H. I.;
Chem. Soc. 2006, 128, 2510−2511. Tieleman, D. P.; Sansom, M. S. Computational Modeling of Realistic
(159) Ono, S.; Naylor, M. R.; Townsend, C. E.; Okumura, C.; Cell Membranes. Chem. Rev. 2019, 119, 6184−6226.
Okada, O.; Lokey, R. S. Conformation and Permeability: Cyclic (177) Lomize, A. L.; Pogozheva, I. D.; Mosberg, H. I. Anisotropic
Hexapeptide Diastereomers. J. Chem. Inf. Model. 2019, 59, 2952− Solvent Model of the Lipid Bilayer. 1. Parameterization of Long-
2963. Range Electrostatics and First Solvation Shell Effects. J. Chem. Inf.
(160) Leung, S. S. F.; Sindhikara, D.; Jacobson, M. P. Simple Model. 2011, 51, 918−929.
Predictive Models of Passive Membrane Permeability Incorporating (178) Leung, S. S. F.; Mijalkovic, J.; Borrelli, K.; Jacobson, M. P.
Size-Dependent Membrane-Water Partition. J. Chem. Inf. Model. Testing Physical Models of Passive Membrane Permeation. J. Chem.
2016, 56, 924−929. Inf. Model. 2012, 52, 1621−1636.

5429 [Link]
J. Chem. Inf. Model. 2023, 63, 5408−5432
Journal of Chemical Information and Modeling [Link]/jcim Perspective

(179) Parisio, G.; Stocchero, M.; Ferrarini, A. Passive Membrane (198) Chawla, N. V.; Bowyer, K. W.; Hall, L. O.; Kegelmeyer, W. P.
Permeability: Beyond the Standard Solubility-Diffusion Model. J. SMOTE: Synthetic Minority Over-sampling Technique. Journal of
Chem. Theory Comput. 2013, 9, 5236−5246. Artificial Intelligence Research 2002, 16, 321−357.
(180) Jakobtorweihen, S.; Zuniga, A. C.; Ingram, T.; Gerlach, T.; (199) Zakharov, A. V.; Peach, M. L.; Sitzmann, M.; Nicklaus, M. C.
Keil, F. J.; Smirnova, I. Predicting solute partitioning in lipid bilayers: QSAR Modeling of Imbalanced High-Throughput Screening Data in
Free energies and partition coefficients from molecular dynamics PubChem. J. Chem. Inf. Model. 2014, 54, 705−712.
simulations and COSMOmic. J. Chem. Phys. 2014, 141, 045102. (200) Cyrus, K.; Wehenkel, M.; Choi, E.-Y.; Han, H.-J.; Lee, H.;
(181) Brocke, S. A.; Degen, A.; MacKerell, A. D.; Dutagaci, B.; Feig, Swanson, H.; Kim, K.-B. Impact of linker length on the activity of
M. Prediction of Membrane Permeation of Drug Molecules by PROTACs. Molecular BioSystems 2011, 7, 359−364.
Combining an Implicit Membrane Model with Machine Learning. J. (201) Weerakoon, D.; Carbajo, R. J.; De Maria, L.; Tyrchan, C.;
Chem. Inf. Model. 2019, 59, 1147−1162. Zhao, H. Impact of PROTAC Linker Plasticity on the Solution
(182) Lomize, A. L.; Pogozheva, I. D. Physics-Based Method for Conformations and Dissociation of the Ternary Complex. J. Chem.
Modeling Passive Membrane Permeability and Translocation Path- Inf. Model. 2022, 62, 340−349.
ways of Bioactive Molecules. J. Chem. Inf. Model. 2019, 59, 3198− (202) Schiedel, M.; Herp, D.; Hammelmann, S.; Swyter, S.;
3213. Lehotzky, A.; Robaa, D.; Oláh, J.; Ovádi, J.; Sippl, W.; Jung, M.
(183) Lomize, A. L.; Hage, J. M.; Schnitzer, K.; Golobokov, K.; Chemically Induced Degradation of Sirtuin 2 (Sirt2) by a Proteolysis
LaFaive, M. B.; Forsyth, A. C.; Pogozheva, I. D. PerMM: A Web Tool Targeting Chimera (PROTAC) Based on Sirtuin Rearranging Ligands
and Database for Analysis of Passive Membrane Permeability and (SirReals). J. Med. Chem. 2018, 61, 482−491.
Translocation Pathways of Bioactive Molecules. J. Chem. Inf. Model. (203) Weng, G.; Li, D.; Kang, Y.; Hou, T. Integrative Modeling of
2019, 59, 3094−3099. PROTAC-Mediated Ternary Complexes. J. Med. Chem. 2021, 64,
(184) Schwobel, J. A. H.; Ebert, A.; Bittermann, K.; Huniar, U.; 16271−16281.
Goss, K.-U.; Klamt, A. COSMOperm: Mechanistic Prediction of (204) Kitchen, D. B.; Decornez, H.; Furr, J. R.; Bajorath, J. Docking
Passive Membrane Permeability for Neutral Compounds and Ions and scoring in virtual screening for drug discovery: methods and
and Its pH Dependence. J. Phys. Chem. B 2020, 124, 3343−3354. applications. Nat. Rev. Drug Discovery 2004, 3, 935−949.
(185) Bittermann, K.; Goss, K.-U. Predicting apparent passive (205) Stanzione, F.; Giangreco, I.; Cole, J. C. Use of molecular
permeability of Caco-2 and MDCK cell-monolayers: A mechanistic docking computational tools in drug discovery. Progress in Medicinal
model. PLoS One 2017, 12, e0190319. Chemistry 2021, 60, 273−343.
(186) Chen, G.; Shen, Z.; Li, Y. A machine-learning-assisted study of (206) Ravindranath, P. A.; Forli, S.; Goodsell, D. S.; Olson, A. J.;
the permeability of small drug-like molecules across lipid membranes. Sanner, M. F. AutoDockFR: Advances in Protein-Ligand Docking
Phys. Chem. Chem. Phys. 2020, 22, 19687−19696. with Explicitly Specified Binding Site Flexibility. PLoS Computational
(187) Pereira, T.; Abbasi, M.; Oliveira, J. L.; Ribeiro, B.; Arrais, J. Biology 2015, 11, e1004586.
Optimizing blood-brain barrier permeation through deep reinforce- (207) Gray, J. J. High-resolution protein-protein docking. Curr.
ment learning for de novo drug design. Bioinformatics 2021, 37, i84− Opin. Struct. Biol. 2006, 16, 183−193.
i92. (208) Drummond, M. L.; Williams, C. I. In Silico Modeling of
(188) de Oliveira, E. C. L.; Santana, K.; Josino, L.; Lima e Lima, A. PROTAC-Mediated Ternary Complexes: Validation and Application.
H.; de Souza de Sales Junior, C. Predicting cell-penetrating peptides J. Chem. Inf. Model. 2019, 59, 1634−1644.
using machine learning algorithms and navigating in their chemical (209) Drummond, M. L.; Henry, A.; Li, H.; Williams, C. I. Improved
space. Sci. Rep. 2021, 11, 7628. Accuracy for Modeling PROTAC-Mediated Ternary Complex
(189) Williams-Noonan, B. J.; Speer, M. N.; Le, T. C.; Sadek, M. M.; Formation and Targeted Protein Degradation via New In Silico
Thompson, P. E.; Norton, R. S.; Yuriev, E.; Barlow, N.; Chalmers, D. Methodologies. J. Chem. Inf. Model. 2020, 60, 5234−5254.
K.; Yarovsky, I. Membrane Permeating Macrocycles: Design Guide- (210) Zaidman, D.; Prilusky, J.; London, N. PRosettaC: Rosetta
lines from Machine Learning. J. Chem. Inf. Model. 2022, 62, 4605− Based Modeling of PROTAC Mediated Ternary Complexes. J. Chem.
4619. Inf. Model. 2020, 60, 4894−4903.
(190) Poongavanam, V.; Kölling, F.; Giese, A.; Göller, A. H.; (211) Bai, N.; Miller, S. A.; Andrianov, G. V.; Yates, M.;
Lehmann, L.; Meibom, D.; Kihlberg, J. Predictive Modeling of Kirubakaran, P.; Karanicolas, J. Rationalizing PROTAC-Mediated
PROTAC Cell Permeability with Machine Learning. ACS Omega Ternary Complex Formation Using Rosetta. J. Chem. Inf. Model. 2021,
2023, 8, 5901−5916. 61, 1368−1382.
(191) Matsson, P.; Doak, B. C.; Over, B.; Kihlberg, J. Cell (212) Pérez-Benito, L.; Henry, A.; Matsoukas, M.-T.; Lopez, L.;
permeability beyond the rule of 5. Adv. Drug Delivery Rev. 2016, 101, Pulido, D.; Royo, M.; Cordomí, A.; Tresadern, G.; Pardo, L. The size
42−61. matters? A computational tool to design bivalent ligands.
(192) David, L.; Wenlock, M.; Barton, P.; Ritzén, A. Prediction of Bioinformatics 2018, 34, 3857−3863.
Chameleonic Efficiency. ChemMedChem. 2021, 16, 2669−2685. (213) Imrie, F.; Bradley, A. R.; van der Schaar, M.; Deane, C. M.
(193) Schwochert, J.; Lao, Y.; Pye, C. R.; Naylor, M. R.; Desai, P. V.; Deep Generative Models for 3D Linker Design. J. Chem. Inf. Model.
Gonzalez Valcarcel, I. C.; Barrett, J. A.; Sawada, G.; Blanco, M.-J.; 2020, 60, 1983−1995.
Lokey, R. S. Stereochemistry Balances Cell Permeability and (214) Yang, Y.; Zheng, S.; Su, S.; Zhao, C.; Xu, J.; Chen, H.
Solubility in the Naturally Derived Phepropeptin Cyclic Peptides. SyntaLinker: automatic fragment linking with deep conditional
ACS Med. Chem. Lett. 2016, 7, 757−761. transformer neural networks. Chemical Science 2020, 11, 8312−8322.
(194) Avdeef, A.; Kansy, M. Can small drugs predict the intrinsic (215) Ignatov, M.; Jindal, A.; Kotelnikov, S.; Beglov, D.; Posternak,
aqueous solubility of ‘beyond Rule of 5′ big drugs? ADMET and G.; Tang, X.; Maisonneuve, P.; Poda, G.; Batey, R. A.; Sicheri, F.;
DMPK 2020, 0, 180−206. Whitty, A.; Tonge, P. J.; Vajda, S.; Kozakov, D. High Accuracy
(195) Palmer, D. S.; Mitchell, J. B. O. Is Experimental Data Quality Prediction of PROTAC Complex Structures. J. Am. Chem. Soc. 2023,
the Limiting Factor in Predicting the Aqueous Solubility of Druglike 145, 7123−7135.
Molecules? Mol. Pharmaceutics 2014, 11, 2962−2972. (216) Li, W.; Zhang, J.; Guo, L.; Wang, Q. Importance of Three-
(196) Meng, J.; Chen, P.; Wahib, M.; Yang, M.; Zheng, L.; Wei, Y.; Body Problems and Protein-Protein Interactions in Proteolysis-
Feng, S.; Liu, W. Boosting the predictive performance with aqueous Targeting Chimera Modeling: Insights from Molecular Dynamics
solubility dataset curation. Scientific Data 2022, 9, 71. Simulations. J. Chem. Inf. Model. 2022, 62, 523−532.
(197) Weng, G.; Cai, X.; Cao, D.; Du, H.; Shen, C.; Deng, Y.; He, (217) Mai, H.; Zimmer, M. H.; Miller, T. F. I. I. I. Exploring
Q.; Yang, B.; Li, D.; Hou, T. PROTAC-DB 2.0: an updated database PROTAC Cooperativity with Coarse-Grained Alchemical Methods. J.
of PROTACs. Nucleic Acids Res. 2023, 51 (D1), D1367−D1372. Phys. Chem. B 2023, 127, 446−455.

5430 [Link]
J. Chem. Inf. Model. 2023, 63, 5408−5432
Journal of Chemical Information and Modeling [Link]/jcim Perspective

(218) Liao, J.; Nie, X.; Unarta, I. C.; Ericksen, S. S.; Tang, W. In (239) Fu, H.; Shao, X.; Cai, W.; Chipot, C. Taming Rugged Free
Silico Modeling and Scoring of PROTAC-Mediated Ternary Energy Landscapes Using an Average Force. Acc. Chem. Res. 2019, 52,
Complex Poses. J. Med. Chem. 2022, 65, 6116−6132. 3254−3264.
(219) Guo, W.-H.; Qi, X.; Yu, X.; Liu, Y.; Chung, C.-I; Bai, F.; Lin, (240) Dellago, C.; Bolhuis, P. G.; Csajka, F. S.; Chandler, D.
X.; Lu, D.; Wang, L.; Chen, J.; et al. Enhancing intracellular Transition path sampling and the calculation of rate constants. J.
accumulation and target engagement of PROTACs with reversible Chem. Phys. 1998, 108, 1964−1977.
covalent chemistry. Nat. Commun. 2020, 11, 4268. (241) Bolhuis, P. G.; Chandler, D.; Dellago, C.; Geissler, P. L.
(220) Shaheer, M.; Singh, R.; Sobhia, M. E. Protein degradation: a Transition Path Sampling: Throwing Ropes Over Rough Mountain
novel computational approach to design protein degrader probes for Passes, in the Dark. Annu. Rev. Phys. Chem. 2002, 53, 291−318.
main protease of SARS-CoV-2. J. Biomol. Struct. Dyn. 2022, 40, (242) Woolf, T. B. Path corrected functionals of stochastic
10905−10917. trajectories: towards relative free energy and reaction coordinate
(221) Yang, Z.; Sun, Y.; Ni, Z.; Yang, C.; Tong, Y.; Liu, Y.; Li, H.; calculations. Chem. Phys. Lett. 1998, 289, 433−441.
Rao, Y. Merging PROTAC and molecular glue for degrading BTK (243) Allen, R. J; Valeriani, C.; Rein ten Wolde, P. Forward flux
and GSPT1 proteins concurrently. Cell Research 2021, 31, 1315− sampling for rare event simulations. J. Phys.: Condens. Matter 2009, 21,
1318. 463102.
(222) Murakami, Y.; Osawa, H.; Kurohara, T.; Yanase, Y.; Ito, T.; (244) Bello-Rivas, J. M.; Elber, R. Exact milestoning. J. Chem. Phys.
Yokoo, H.; Shibata, N.; Naito, M.; Aritake, K.; Demizu, Y. Structure− 2015, 142, 94102.
activity relationship study of PROTACs against hematopoietic (245) Huber, G. A.; Kim, S. Weighted-ensemble Brownian dynamics
prostaglandin D2 synthase. RSC Med. Chem. 2022, 13, 1495−1503. simulations for protein association reactions. Biophys. J. 1996, 70, 97−
(223) Xue, X.; Kang, J.-B.; Yang, X.; Li, N.; Chang, L.; Ji, J.; Meng, 110.
X.-K.; Zhang, H.-Q.; Zhong, Y.; Yu, S.-P.; Wu, W.-Y.; Wang, X.-L.; Li, (246) Zuckerman, D. M.; Chong, L. T. Weighted Ensemble
N.-G.; Sun, S.-L. An efficient strategy for digging protein-protein Simulation: Review of Methodology, Applications, and Software.
interactions for rational drug design - A case study with HIF-1α/VHL. Annual Review of Biophysics 2017, 46, 43−57.
European journal of medicinal chemistry 2022, 227, 113871. (247) Ray, D.; Stone, S. E.; Andricioaei, I. Markovian Weighted
(224) Carlson, H. A. Protein flexibility and drug design: how to hit a Ensemble Milestoning (M-WEM): Long-Time Kinetics from Short
moving target. Curr. Opin. Chem. Biol. 2002, 6, 447−452. Trajectories. J. Chem. Theory Comput. 2022, 18, 79−95.
(225) Cozzini, P.; Kellogg, G. E.; Spyrakis, F.; Abraham, D. J.; (248) Swendsen, R. H.; Wang, J.-S. Replica Monte Carlo Simulation
Costantino, G.; Emerson, A.; Fanelli, F.; Gohlke, H.; Kuhn, L. A.; of Spin-Glasses. Phys. Rev. Lett. 1986, 57, 2607−2609.
Morris, G. M.; Orozco, M.; Pertinhez, T. A.; Rizzi, M.; Sotriffer, C. A. (249) Sugita, Y.; Okamoto, Y. Replica-exchange molecular dynamics
Target Flexibility: An Emerging Consideration in Drug Discovery and method for protein folding. Chem. Phys. Lett. 1999, 314, 141−151.
Design. J. Med. Chem. 2008, 51, 6237−6255. (250) Wang, L.; Friesner, R. A.; Berne, B. J. Replica Exchange with
(226) Torrie, G. M.; Valleau, J. P. Nonphysical sampling Solute Scaling: A More Efficient Version of Replica Exchange with
distributions in Monte Carlo free-energy estimation: Umbrella Solute Tempering (REST2). J. Phys. Chem. B 2011, 115, 9431−9438.
sampling. J. Comput. Phys. 1977, 23, 187−199. (251) Moors, S. L. C.; Michielssens, S.; Ceulemans, A. Improved
(227) Isralewitz, B.; Gao, M.; Schulten, K. Steered molecular Replica Exchange Method for Native-State Protein Sampling. J. Chem.
dynamics and mechanical functions of proteins. Curr. Opin. Struct. Theory Comput. 2011, 7, 231−237.
Biol. 2001, 11, 224−230. (252) Roitberg, A. E.; Okur, A.; Simmerling, C. Coupling of Replica
(228) Hamelberg, D.; Mongan, J.; McCammon, J. A. Accelerated Exchange Simulations to a Non-Boltzmann Structure Reservoir. J.
molecular dynamics: A promising and efficient simulation method for Phys. Chem. B 2007, 111, 2415−2418.
biomolecules. J. Chem. Phys. 2004, 120, 11919−11929. (253) Kasavajhala, K.; Lam, K.; Simmerling, C. Exploring Protocols
(229) Marchi, M.; Ballone, P. Adiabatic bias molecular dynamics: A to Build Reservoirs to Accelerate Temperature Replica Exchange MD
method to navigate the conformational space of complex molecular Simulations. J. Chem. Theory Comput. 2020, 16, 7776−7799.
systems. J. Chem. Phys. 1999, 110, 3697−3702. (254) Copperman, J.; Zuckerman, D. M. Accelerated Estimation of
(230) Darve, E.; Pohorille, A. Calculating free energies using average Long-Timescale Kinetics from Weighted Ensemble Simulation via
force. J. Chem. Phys. 2001, 115, 9169−9183. Non-Markovian “Microbin” Analysis. J. Chem. Theory Comput. 2020,
(231) Comer, J.; Gumbart, J. C.; Hénin, J.; Lelièvre, T.; Pohorille, 16, 6763−6775.
A.; Chipot, C. The Adaptive Biasing Force Method: Everything You (255) Ojha, A. A.; Thakur, S.; Ahn, S.-H.; Amaro, R. E. DeepWEST:
Always Wanted To Know but Were Afraid To Ask. J. Phys. Chem. B Deep Learning of Kinetic Models with the Weighted Ensemble
2015, 119, 1129−1151. Simulation Toolkit for Enhanced Sampling. J. Chem. Theory Comput.
(232) Laio, A.; Parrinello, M. Escaping free-energy minima. Proc. 2023, 19, 1342−1359.
Natl. Acad. Sci. U. S. A. 2002, 99, 12562−12566. (256) Buchete, N.-V.; Hummer, G. Coarse Master Equations for
(233) Laio, A.; Rodriguez-Fortea, A.; Gervasio, F. L.; Ceccarelli, M.; Peptide Folding Dynamics. J. Phys. Chem. B 2008, 112, 6057−6069.
Parrinello, M. Assessing the Accuracy of Metadynamics. J. Phys. Chem. (257) Pande, V. S.; Beauchamp, K.; Bowman, G. R. Everything you
B 2005, 109, 6714−6721. wanted to know about Markov State Models but were afraid to ask.
(234) Cavalli, A.; Spitaleri, A.; Saladino, G.; Gervasio, F. L. Methods 2010, 52, 99−105.
Investigating Drug−Target Association and Dissociation Mechanisms (258) Chodera, J. D.; Noé, F. Markov state models of biomolecular
Using Metadynamics-Based Algorithms. Acc. Chem. Res. 2015, 48, conformational dynamics. Curr. Opin. Struct. Biol. 2014, 25, 135−144.
277−285. (259) Bowman, G. R.; Pande, V. S.; Noé, F., Eds. An Introduction to
(235) Barducci, A.; Bussi, G.; Parrinello, M. Well-Tempered Markov State Models and Their Application to Long Timescale Molecular
Metadynamics: A Smoothly Converging and Tunable Free-Energy Simulation; Springer: Netherlands, 2014; Vol. 797.
Method. Phys. Rev. Lett. 2008, 100, 20603. (260) Wang, W.; Cao, S.; Zhu, L.; Huang, X. Constructing Markov
(236) Dama, J. F.; Parrinello, M.; Voth, G. A. Well-Tempered State Models to elucidate the functional conformational changes of
Metadynamics Converges Asymptotically. Phys. Rev. Lett. 2014, 112, complex biomolecules. WIREs Computational Molecular Science 2018,
240602. 8, e1343.
(237) Piana, S.; Laio, A. A Bias-Exchange Approach to Protein (261) Wang, S.; Sun, S.; Li, Z.; Zhang, R.; Xu, J. Accurate De Novo
Folding. J. Phys. Chem. B 2007, 111, 4553−4559. Prediction of Protein Contact Map by Ultra-Deep Learning Model.
(238) Fu, H.; Zhang, H.; Chen, H.; Shao, X.; Chipot, C.; Cai, W. PLOS Computational Biology 2017, 13, e1005324.
Zooming across the Free-Energy Landscape: Shaving Barriers, and (262) Senior, A. W.; et al. Improved protein structure prediction
Flooding Valleys. J. Phys. Chem. Lett. 2018, 9, 4738−4745. using potentials from deep learning. Nature 2020, 577, 706−710.

5431 [Link]
J. Chem. Inf. Model. 2023, 63, 5408−5432
Journal of Chemical Information and Modeling [Link]/jcim Perspective

(263) Pearce, R.; Zhang, Y. Deep learning techniques have AUTOTAC chemical biology platform for targeted protein
significantly impacted protein structure prediction and protein design. degradation via the autophagy-lysosome system. Nat. Commun.
Curr. Opin. Struct. Biol. 2021, 68, 194−207. 2022, 13, 904.
(264) Jumper, J.; et al. Highly accurate protein structure prediction (284) Henning, N. J.; et al. Deubiquitinase-targeting chimeras for
with AlphaFold. Nature 2021, 596, 583−589. targeted protein stabilization. Nat. Chem. Biol. 2022, 18, 412−421.
(265) Strokach, A.; Kim, P. M. Deep generative modeling for protein (285) Foley, K.; Dai, Y.; Ding, Q.; Du, F.; Li, J.; Lv, C.; Prince, T.;
design. Curr. Opin. Struct. Biol. 2022, 72, 226−236. Sun, Y.; Wang, M.; Wang, R.; et al. Tumor-selective, chaperone-
(266) Park, D.; Izaguirre, J.; Coffey, R.; Xu, H. Modeling the effect mediated protein degradation (CHAMP) of the bromodomain
of cooperativity in ternary complex formation and targeted protein transcription factor BRD4. Eur. J. Cancer 2022, 174, S27.
degradation mediated by heterobifunctional degraders. ACS Bio & (286) Siriwardena, S. U.; Munkanatta Godage, D. N.; Shoba, V. M.;
Med. Chem. Au 2022, 3 (1), 74−86. Lai, S.; Shi, M.; Wu, P.; Chaudhary, S. K.; Schreiber, S. L.;
(267) Haid, R. T. U.; Reichel, A. A Mechanistic Pharmacodynamic Choudhary, A. Phosphorylation-inducing chimeric small molecules. J.
Modeling Framework for the Assessment and Optimization of Am. Chem. Soc. 2020, 142, 14052−14057.
Proteolysis Targeting Chimeras (PROTACs). Pharmaceutics 2023,
15, 195.
(268) Douglass, E. F.; Miller, C. J. A comprehensive kinetic model
for ternary complex catalysis. ChemRxiv. Version 2. 2022.
(269) Johnson, K. A.; Goody, R. S. The original Michaelis constant:
translation of the 1913 Michaelis-Menten paper. Biochemistry 2011,
50, 8264−8269.
(270) Watt, G. F.; Scott-Stevens, P.; Gaohua, L. Targeted protein
degradation in vivo with Proteolysis Targeting Chimeras: Current
status and future considerations. Drug Discovery Today: Technologies
2019, 31, 69−80.
(271) Paiva, S.-L.; Crews, C. M. Targeted protein degradation:
elements of PROTAC design. Curr. Opin. Chem. Biol. 2019, 50, 111−
119.
(272) Chan, K.-H.; Zengerle, M.; Testa, A.; Ciulli, A. Impact of
Target Warhead and Linkage Vector on Inducing Protein
Degradation: Comparison of Bromodomain and Extra-Terminal
(BET) Degraders Derived from Triazolodiazepine (JQ1) and
Tetrahydroquinoline (I-BET726) BET Inhibitor Scaffolds. Journal of
medicinal chemistry 2018, 61, 504−513.
(273) Jiang, B.; et al. Discovery and resistance mechanism of a
selective CDK12 degrader. Nat. Chem. Biol. 2021, 17, 675−683.
(274) Bartlett, D. W.; Gilbert, A. M. A kinetic proofreading model
for bispecific protein degraders. Journal of pharmacokinetics and
pharmacodynamics 2021, 48, 149−163.
(275) Desantis, J.; Mammoli, A.; Eleuteri, M.; Coletti, A.; Croci, F.;
Macchiarulo, A.; Goracci, L. PROTACs bearing piperazine-containing
linkers: what effect on their protonation state? RSC Adv. 2022, 12,
21968−21977.
(276) Schwantes, C. R.; Pande, V. S. Improvements in Markov State
Model Construction Reveal Many Non-Native Interactions in the
Folding of NTL9. J. Chem. Theory Comput. 2013, 9, 2000−2009.
(277) Donyapour, N.; Roussey, N. M.; Dickson, A. REVO:
Resampling of ensembles by variation optimization. J. Chem. Phys.
2019, 150, 244112.
(278) Shaw, D. E.; Adams, P. J.; Azaria, A.; Bank, J. A.; Batson, B.;
Bell, A.; Bergdorf, M.; Bhatt, J.; Butts, J. A.; Correia, T.; et al. Anton 3:
twenty microseconds of molecular dynamics simulation before lunch.
Proceedings of the International Conference for High Performance
Computing, Networking, Storage and Analysis 2021, 1−11.
(279) Larson, S. M.; Snow, C. D.; Shirts, M.; Pande, V. S. Folding@
Home and Genome@Home: Using distributed computing to tackle
previously intractable problems in computational biologyarXiv.
arXiv:0901.0866v1, Version 1. 2009.
(280) Lu, D.; Yu, X.; Lin, H.; Cheng, R.; Monroy, E. Y.; Qi, X.;
Wang, M. C.; Wang, J. Applications of covalent chemistry in targeted
protein degradation. Chem. Soc. Rev. 2022, 51, 9243−9261.
(281) Ahn, G.; Banik, S. M.; Miller, C. L.; Riley, N. M.; Cochran, J.
R.; Bertozzi, C. R. LYTACs that engage the asialoglycoprotein
receptor for targeted protein degradation. Nat. Chem. Biol. 2021, 17,
937−946.
(282) Alabi, S. B.; Crews, C. M. Major advances in targeted protein
degradation: PROTACs, LYTACs, and MADTACs. J. Biol. Chem.
2021, 296, 100647.
(283) Ji, C. H.; Kim, H. Y.; Lee, M. J.; Heo, A. J.; Park, D. Y.; Lim,
S.; Shin, S.; Ganipisetti, S.; Yang, W. S.; Jung, C. A.; et al. The

5432 [Link]
J. Chem. Inf. Model. 2023, 63, 5408−5432