Review

[Link] pISSN 1738-1088 / eISSN 2093-4327

Clinical Psychopharmacology and Neuroscience 2022;20(1):26-36 Copyrightⓒ 2022, Korean College of Neuropsychopharmacology

Computational Neuroscience Approach to Psychiatry:

A Review on Theory-driven Approaches
Ali Khaleghi1, Mohammad Reza Mohammadi1, Kian Shahi1, Ali Motie Nasrabadi2
1 2
Psychiatry and Psychology Research Center, Tehran University of Medical Sciences, Department of Biomedical Engineering, Shahed University,
Tehran, Iran

Translating progress in neuroscience into clinical benefits for patients with psychiatric disorders is challenging because
it involves the brain as the most complex organ and its interaction with a complex environment and condition. Dealing
with such complexity requires powerful techniques. Computational neuroscience approach to psychiatry integrates mul-
tiple levels and types of simulation, analysis and computation according to the different types of computational models
to enhance comprehending, prediction and treatment of psychiatric disorder. This approach comprises two approaches:
theory-driven and data-driven. In this review, we focus on recent advances in theory-driven approaches that mathemati-
cally and mechanistically examine the relationships between disorder-related changes and behavior at different level
of brain organization. We discuss recent progresses in computational neuroscience models that relate to psychiatry
and show how principles of neural computational modeling can be employed to explain psychopathology.
KEY WORDS: Neuroscience; Theoretical models; Psychiatry; Mathematical computing.

INTRODUCTION similar behavioral disturbances and, conversely, one bio-

logical dysfunction may affect different apparently un-
Research in the field of psychiatry has experienced a related psychological functions [1-3]. Generally, in-
stagnation due to the slow translation of neuroscience ad- complete understanding of human cognitive phenotypes
vances to definitive measures for the recovery of patients has led to the lack of a concrete bridge between the phe-
with mental disorders. The large complexity of the human nomenological and the molecular events. As a result, we
brain and the problems in recognizing and explaining the see frequent concerns and questions about the classi-
biological, contextual and environmental underpinnings fication of psychiatric disorders, and as a clear example,
of mental functions astonish the study of the pathophysi- the ongoing revisions of the Diagnostic and Statistical
ology and etiology of mental disorders, and make an ex- Manual of Mental Disorders (DSM) of the American
planatory gap between neuroscience and biological psy- Psychiatric Association [4-6]. It is very difficult to deal
chiatry at intermediate levels of description that must bind with these challenges without theoretical models, con-
ideas and hypotheses at the molecular level to those man- ceptual frameworks and powerful computational tools.
ifested at the level of clinical entities, such as depression Although experimental studies yield valuable and useful
and schizophrenia. On the other hand, the corre- information to understand the cognitive processes and bi-
spondence between these levels is not one to one. In other ological disturbances in mental illness [7-11], they are
words, different biological dysfunctions may generate usually limited by practical, economic or ethical factors.
In recent years, computational neuroscience has shown
Received: April 13, 2021 / Revised: June 9, 2021 great promise in the field of psychiatry research and com-
Accepted: June 14, 2021 putational psychiatry by establishing a connection be-
Address for correspondence: Ali Khaleghi tween pathophysiological and phenomenological aspects
Psychiatry and Psychology Research Center, Tehran University of
Medical Sciences, South Kargar Av., Tehran 1333795914, Iran of mental illness, thereby reshaping current nosology in
E-mail: alikhaleghi_bme84@[Link] more biologically dimensions. Computational psychiatry
ORCID: [Link]
This is an Open-Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License ([Link]
which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.

26
 Computational Neuroscience Approach to Psychiatry 27

is an emerging field in expansion at the junction between able relationships between variables (including observable
psychiatry and computational neuroscience [12-14]. variables or hypothetical ones), and data-driven ap-
Computational neuroscience is a subfield of neuro- proaches that utilize various statistical models and ma-
science which utilizes theoretical analysis, mathematical chine learning techniques for the analysis of empirical
models and abstractions of the brain to perceive the prin- and experimental data to provide a suitable solution for
ciples that control development, structure, physiology, the issues in the classification, forecasting and treatment
emotional and cognitive processes of the nervous system of diseases. Data-driven approaches apply machine
[15]. The goal of mathematical models in computational learning techniques to high dimensional data to improve
neuroscience is to capture the main properties of a bio- diseases classification, predict the outcomes of treatment
logical system at multiple spatiotemporal scales, from and improve treatment selection. Data-driven approaches
membrane potentials, neurotransmitter function and are often used to develop clinically useful applications,
topographic architecture to a psychological faculty like while theory-driven approaches are often used to en-
behavior or learning [16,17]. These computational mod- hance our insights into disorders and diseases. Several re-
els frame ideas that can be directly examined by psycho- cent studies have thoroughly reviewed data-driven ap-
logical or biological experiments. While psychiatric dis- proaches in detail and examined various aspects of these
orders are described essentially by high-level symptoms, models [12,14,19-21]. However, much less attention has
computational models can help frame symptoms and hy- been paid to theory-driven approaches, which can be an
potheses to fill the gap between psychiatry and neuro- appropriate complement to data-driven methods.
biology. For instance, Maia and Frank showed how mod- Therefore, in this review, we focus on recent advances in
eling using an abductive or deductive approach can result theory-driven approaches that mathematically and mech-
in different predictions for psychiatry [18]. Using the ab- anistically examine the relationships between disorder-
ductive approach researchers start from the proposition of related changes and behavior at different level of brain
a model of normal function and change the model in dif- organization.
ferent ways to produce new hypotheses of brain and bio-
logical dysfunction. Then, these models are fitted to the COMPUTATIONAL MODELING
behavior of patients to determine which hypothesis can
explain the patients’ performance. The top hypothesis can Mathematical or biophysical models provide a quanti-
then be revised in a try to illustrate the deficits at lower tative expression for biological phenomena. These syn-
levels, or used to plan new experimental tests that will ex- thetic models are built based on physical and mathemat-
amine the dysfunction proposed by the hypothesis. On ical rules, computer-aided tools and biological data from
the other hand, the deductive approach starts from the different sources related to the system of interest (e.g., a
proposition of known neurobiological deficits reported in neural system) and investigate the interaction between
mental illness, and implement these deficiencies in a components through simulations and mathematical anal-
computational model. Then, the performance of patients ysis [16]. Figure 1 shows the research process involved in
is compared to those of the model. If the model can ex- computational modeling to study different biological
plain the performance and function deficits reported in systems. Based on goal, methodology, simulation level
patients, it provides a reasonable mechanistic platform and the capacity to imitate mental illness, three types of
that link biology and neurobiology to behavior. Therefore, models can be identified in computational neuroscience
computational neuroscience and computational psychia- and psychiatry: the phenotypic model, the neuro-
try are important to the future of psychiatry research and biological model, and the intermediate model. Scientists
will probably play a main role in the logical development choose one of these types of models depending on how
of nosologies, preventive and treatment strategies. nearly and in what sense they want to imitate the target to
On the other hand, there are two very common ap- achieve an appropriate simulated object. In fact, the main
proaches in computational psychiatry that can be the difference between these models is the gap between the
bridge between neuroscience and psychiatry: theory- low level biological function with a high level phenom-
driven approaches that look for mechanistically explain- enological manifestation, i.e., apparent thought, emotion
28 A. Khaleghi, et al.

Fig. 1. Computational modeling re-

search process to simulate and study
biological systems. There may be
direct flows from biology to compu-
tational simulation, but in this process
we do need powerful hardware and
software resources, robotics and other
technology-related resources. There-
fore, considering the role of tech-
nology in this process and flow dia-
gram seems logical.

and behavior. gree to which they represent a brain like activity. Thus,
Psychiatric disorders are definitively correlated with there is a range of this type of model from machine learn-
brain pathologies, but, as mentioned above, they are cur- ing and artificial intelligence systems that focus merely on
rently characterized and categorized by the DSM based the simulation of psychiatric symptoms to a more bio-
on their phenotype manifestations. These disorders in- logically cognitive system that can produce useful activ-
volve a broad spectrum of expressions and causes, rang- ities in coarsely brain like ways based on high level archi-
ing from largely ambiguous biological and environmental tecture of the brain.
bases to observable behavior and subjective experience Computer simulations of psychiatric disorders in the
[21]. Each of those three models focuses on a particular form of phenotypic models may help us to better under-
aspect of psychiatric disorders, based on explicit or im- stand the characteristic verbal and emotional responses of
plicit, theoretical or practical assumptions that will be ex- a patient who suffers from a specific mental illness and
plored below. provide a simulator to investigate potential reactions to
various stimuli [22]. It may even be possible to under-
THE PHENOTYPIC MODELS stand the fundamental relationship between specific
symptoms and the underlying neurobiological function
This types of models attempts to imitate clinical pheno- that correlates with or even causes these symptoms. For
types or high level manifestations of the disorder. Turing’s example, Akbari and Zhuiykov [23] recently developed a
approach to brain simulation is consistent with this view. bioinspired optoelectronically engineered neurorobotics
The focus on mental disorder simulation here is the be- model with sensorimotor functionalities. The neural en-
havioral phenotype. An ideal phenotypic model should gineering framework captures a high level illustration of a
go beyond a black box and be able to simulate a real life cognitive theory and integrates it with related neuro-
system with the ability to provide somatic responses and physiological and anatomical constraints, generating a
duplicate verbal and physiological like characteristics of detailed mechanistic model of how numerous interacting
psychiatric disorders. Phenotypic models differ in the de- neurons can produce the desired behavior efficiently. In
 Computational Neuroscience Approach to Psychiatry 29

Fig. 2. Simplified phenotypic modeling process of the neurorobotics platform. The user can develop a neurorobotic experiment through a brain
model linked to a robotic body or neurobot that interacts in the dynamic environment. Then, the experiment is triggered using a synchronized neural
physics simulation. Finally, the performance can be displayed and evaluated in an interactive manner. The brain picture was reused from the article
of Eliasmith et al. (Science 2012;338:1202-1205) [25] with original copyright holder’s permission under license number 5084650890717.

[24], a neurorobotics platform is introduced that connects logical factors rather than looking at diseases merely as
artificial brain neural networks with robots in a compre- sets of symptoms. The scope of this view would lead us to
hensive simulation model. This platform can behave sim- design and develop therapies to selectively target the un-
ilarly to humans in response to sensorimotor stimuli (Fig. derlying abnormal mechanisms. In fact, we can embed a
2). Semantic pointer architecture unified network (Spaun) specific mutation into the neurobiological model and
is a neuron model of the brain that produces motor func- then study how that change affects any part of the bio-
tions by a mechanistically modeled arm in response to logical processes [29]. These types of models are suitable
visual stimuli [25]. In a more clinical framework, some for testing hypotheses. They can test possible pathological
studies introduced interactive robots to recognize im- mechanisms at different low levels. In addition, these
pulsive or inattentive behaviors in patients with Attention models can be used to investigate the effects of various
Deficit Hyperactivity Disorder (ADHD) and autism and drugs and therapies on synapses and neurotransmitters
provide immediate feedback to them [26,27]. As a result, and other low-level biological processes. However, some
a phenotypic model of mental illness may imitate and rec- researchers have raised doubts regarding the capability of
ognize the cognitive processes, perception and reactions such modeling approach to generate intelligent actions
involved in a specific disorder. However, the question like a human brain. Among the various practical and con-
that arises in the mind is whether the computer can real- ceptual concerns, the most serious one in the context of
istically mimic the real life behavior of a patient suffering psychiatric disorders relates to the controversial capacity
from a mental illness. In fact, this type of simulation could of this approach to reliably simulate psychiatric symptoms.
not be realistically linked to the functioning of the brain or In addition, most of these models relate low-level neuro-
even biological structures. Therefore, the phenotypic type biological processes to higher-level neurobiological proc-
of the model provides no new information about the po- esses in a mental illness rather than behavioral manifes-
tential role of biological factors in the pathogenesis of tations of that disease. For instance, Tsoutsouras et al. [30]
mental illness, nor about the therapeutic effect of certain presented a modeling approach based on cellular autom-
biological interventions. ata for the loss of neurons and active synapses in limbic
brain areas (epileptogenic focus) that causes epileptic
THE NEUROBIOLOGICAL MODELS seizures by propagating the pathological dynamics from
the focal to healthy brain regions. Eventually, they were
This type of models attempts to imitate low level brain able to simulate the electrical activity or electro-
function through the real molecular, synaptic, connec- encephalogram of epileptic brain (Fig. 3). In addition, a
tivity and neuronal characteristics of the brain in mental recent study reviewed neurobiological models of psycho-
illness. The underlying hypothesis in neurobiological models therapy and explained the important role of neuro-
is that high level phenomena are caused by low level neu- plasticity models as the neurobiological basis of psycho-
robiological activity. In fact, these models follow a bot- therapy in mechanistic understanding of underlying neu-
tom-up approach to explain a potential disease mecha- ral mechanisms involved in this therapeutic approach
nism in biology [28]. These neurobiological models [31]. Zilcha-Mano et al. [32] explained the neuro-
would help us to recategorize disorders based on bio- biological underpinnings of therapeutic expectancy and
30 A. Khaleghi, et al.

Fig. 3. A simple example of the neurobiological modeling process of epilepsy that ultimately results in the simulation of epileptic EEG signal
compared to normal EEG signal.
EEG, electroencephalogram.

alliance and underlined the importance of neurobiological neurotransmitters could result in neurodynamics or be-
models to understand these effects. Iglesias et al. [33] ex- havioral disturbances) or deductively to infer the possible
plained how neurobiological models can be used to study consequences of suspected or known causes (for in-
the effects of neuromodulatory approaches in computa- stance, what effects a variation in the concentration of a
tional psychiatry. certain neurotransmitter would have on neural-circuit or
behavior dynamics). Previous studies have been used var-
THE INTERMEDIATE MODELS ious synthetic biophysical models to connect neuro-
biological abnormalities in mental illness to its neuro-
These types of models have received the most attention dynamical and behavioral consequences [34]. Cortical
in the field of computational psychiatry. The intermediate pyramidal neurons and GABAergic interneurons form an
models stand in between the neurobiological and the important class of these models that has provided val-
phenotypic models. These types of models attempt to es- uable insights in psychiatry. Reducing NMDAR density
tablish a logical relationship between high-level psychi- within inhibitory interneurons suggested that working
atric symptoms and low-level biological dysfunction. memory in schizophrenia must be sensitive to distractors
They differ from both the neurobiological and the pheno- similar to the objects held in working memory [35].
typic models, because they neither try to mimic symptoms Different integrative models have been presented to dis-
realistically nor biological structures. Rather, these mod- cover the relationship between NMDAR density and
els would provide means and frameworks for improved BOLD signals. They suggested that ketamine causes psy-
theoretical comprehending of mental disorders. Figure 4 chosis symptoms and destroys the negative relation be-
shows an example of the intermediate model for con- tween task-related modes and the resting-state default
ceptualization and understanding of schizophrenia based mode [36,37]. Furthermore, dysfunction in the posterior
on the dopamine hypothesis of schizophrenia. Various regions of the brain may cause impairments in attentional
models have been introduced using this approach that orientation on different mental arithmetic tasks in ADHD
can be categorized into three subtypes: synthetic, algo- [7]. In another study, a neural network model explored
rithmic and Bayesian models. the effect of serotonergic and glutamatergic disturbances
Synthetic models usually connect different levels of in obsessive-compulsive disorder, and suggested that both
analysis and simulation, and can be used abductively to increased glutamate and decreased serotonin can be a
infer the possible causes for a known outcome (for in- likely neural substrate for obsession [1]. Similar con-
stance, what kind of perturbation in the release of some nections from synaptic activities to high-level functions
 Computational Neuroscience Approach to Psychiatry 31

Fig. 4. An example of the intermediate model for conceptualization and understanding of schizophrenia based on the dopamine hypothesis of
schizophrenia. According to this model, schizophrenia is best conceived as a complex disorder which involves multiple dopaminergic pathways.

have been achieved through biophysically realistic de- Such behaviors are related to subject’s survival and shape
tailed models of the cortico-striata-thalamic circuits [38,39]. the main concern of neural circuits. According to the role
These models explained different aspects of schizo- of dopamine in modulating NMDAR plasticity and the im-
phrenia, Tourette’s syndrome, addiction and Parkinson’s portance of prediction errors for associative learning, it
disease [18,40]. In summary, synthetic models provide an has suggested that phasic dopamine release under pre-
analytical tool to bridge biological details to symptoms diction error encoding may serve as a reference to adjust
through understanding causally complex relationships synaptic connections and functions in task-relevant cir-
between levels of analysis. However, it should be noted cuits, and that abnormal dopaminergic activities may
that due to the simplifications considered in these models, cause maladaptive learning [33]. However, many studies
conclusions are limited to the levels of analysis involved have suggested that dopamine also encodes uncertainty
in the model. or precision [45-47]. The field of RL has drawn two differ-
Algorithmic reinforcement learning (RL) models pro- ent approaches in which past experience and prior beliefs
vide a quantitative explanation of the mechanisms under- are utilized to predict future reward and punishment:
lying disturbances of executive function, emotion, deci- model-based and model-free cognition (a detailed de-
sion-making and other functions. These powerful frame- scription of these approaches is beyond the scope of the
works have formed many aspects and details of systems review) [48]. These notions have guided concepts and re-
neuroscience over the last decade. They provide a norma- searches on various mental disorders. Previous studies
tive model of choice behavior with neural substrates and have linked multiple aspects of RL to a wide variety range
neurobiological correlates [41]. Although they differ from of symptoms of psychiatric diseases, including: positive
synthetic models, they can characterize different aspects symptoms in schizophrenia (e.g., hallucinations and delu-
of neural activity and subsequent behavior [42,43]. sions) with atypical reward prediction errors [49,50], ab-
Perhaps the most popular correlate is the relationship be- normal activity in the limbic motivational systems in re-
tween phasic dopaminergic activity in the basal ganglia sponse to neural stimuli [51], and aberrant levels of in-
and reward-seeking/punishment-avoiding behaviors in centive salience because of abnormally increased levels
the presence of reward and punishment as reinforces [44]. of dopaminergic neurotransmission in prefrontal cortex
32 A. Khaleghi, et al.

and ventral striatum [52,53]; negative symptoms of schiz- plications in the computational psychiatry. Our goal here
ophrenia with aberrant effort-cost computations and in- was not to focus on psychiatric disorders such as depres-
ability for representing the expected value of rewards sion, autism, ADHD, anxiety and others, but rather on
[54-57]; relapse prediction in alcohol-dependence with computational modeling approaches in neuroscience to
conditioned stimulus-related activation in the nucleus ac- characterize and measure the pathobiological mecha-
cumbens [58]; impulsivity in ADHD with reduced delay nisms involved in psychiatric symptoms. Discussion of
aversion due to over-discounting of delayed rewards mental illness modeling should involve the important
[59,60]; anhedonia in depression related to a deficit in re- question of the purposes of doing so. Actually, there is a
ward sensitivity in a different way from that in dop- critical need for new clinical tests that identify the main
aminergic transmission affecting learning [61]. abnormality that probably underlies the symptoms of a
An important advantage of characterizing reward func- patient such that targeted treatment and medication
tions in terms of prior beliefs or past experiences is that the would be selected as well as spectrum diseases would be
beliefs, that are responsible for individual behavior, can split into more distinct subgroups mechanistically. This is
be defined, enabling computational and quantitative phe- where we need to look for answers. In other words, the
notyping in terms of attitudes and beliefs [62,63]. This ap- long-term objective of neuroscience-based computa-
proach shapes the principle of many current computa- tional modeling in psychiatry should be to address this
tional psychiatry innovations, and even has been ex- critical need, i.e., better understanding of the pathophy-
tended to game-theoretic models of interpersonal ex- siological processes causing mental disorders, and ulti-
change that involves strategic cooperation, competition mately diagnosing and treating these pathologies more
and social fairness [64,65]. Stag-hunt game and mul- effectively. The success of most scientific attempts de-
ti-round trust game are two prominent examples of these pends on the accuracy and preciseness with which phe-
models that have revealed disorder-specific play styles in nomena can be mechanistically modeled. In psychiatry,
many psychiatric disorders, including autism, ADHD, models are usually developed by prose according to a
schizophrenia, depression and borderline personality dis- synthesis and fusion of available experimental findings
orders [64,66]. These Bayesian models can be employed [72]. An important point in computational modeling is
to better understand the nature of problem and its that these models should not be formulated to test certain
solution. One important property of Bayesian models is hypotheses about brain function and dysfunction. Instead,
their emphasis on the use and representation of un- it is much better that they view as mathematical ob-
certainty [67]. Accordingly, it has been shown that the sta- servatories to examine and validate different candidate
tistics of aversive experience play an important role in hypotheses. Thus, a computational model is a powerful
several processes, from learned helplessness and depres- tool, not only a hypothesis, and can be linked to clinical
sion to familiarity in fear conditioning [68-70]. Bayesian tools for patient evaluation. To this end, we need to take
models can also be utilized to explore a possible relation- advantages of each of above computational models and
ship between a certain symptom and suboptimal inference. minimize their limitations and complexities to achieve a
For instance, subjects with high trait anxiety cannot up- comprehensive applicable model. In fact, the complexity
date optimally how volatile an aversive situation is, while of mental disorders and the current biological-pheno-
low anxiety controls present close to Bayes-optimum be- menological gap can benefit from a multi-level method
havior [71]. Generally, these extensions and other ap- that merges multiple modes of simulation and modeling: a
proaches derived from behavioral economics could pro- top-down high-level phenotypically symptom oriented
vide a unique opportunity to describe different psycho- mode and a bottom-up low-level biologically oriented
pathies and their genetic or neural correlates. mode. Successful modeling of mental illness should in-
clude these two modes that correlate with the bottommost
DISCUSSION and uppermost layers of brain functioning. The pheno-
typically symptom oriented modeling can benefit from
In this brief overview, we have reviewed theory-driven the artificial intelligence, whereas neurobiological mod-
approaches in computational neuroscience and their ap- eling can benefit from the different methodologies of
 Computational Neuroscience Approach to Psychiatry 33

computational neuroscience. In computational neuro- can be supported by a more focus on establishing utility
science, researcher must learn to compute and analyze through the active application of computational appro-
without knowing all the parameters. In other words, the aches in clinical trials. Combining theory- and data-driv-
model parameters characterizing brain networks are un- en approaches can be an appropriate way [14,75-79].
certain, numerous and can change over time or situation. Combining theory- and data-driven approaches can be
The effects of the uncertain parameters of the model and very helpful from an applied point of view [80]. Theory-
the uncertainty in model predictions can be studied driven approaches provide an estimation of certain fea-
through repeated evaluations of the model, which is often tures relevant to a specific disorder, if the prior knowledge
computationally demanding [73]. and mechanistic understanding of the condition is suffi-
With the development of numerical algorithms and ciently accurate. As a result, they substantially reduce the
computational and computer hardware, we will probably data dimensionality by limiting the data set to the sub-
include more phenotypic and even neurobiological mod- space of a few relevant features and parameters. Then, da-
els in clinical practice in the future. However, according ta-driven approaches can apply machine learning ap-
to the characteristics, properties, advantages and limi- proaches to this lower dimensional data set with higher
tations of each category of above models, it seems that the reliability and efficiency. Such an approach has already
intermediate models are now more usable for better un- been used in a few studies to investigate Huntington’s [81]
derstanding of mental illness and clinical setting in general. and schizophrenia [82], and has led to considerable im-
Regarding the subtypes of intermediate models, it should provements in the classification of behaviors and sub-
be noted that the distinction between synthetic, algo- types of these diseases.
rithmic and Bayesian models can be vague. For instance,
a biophysically detailed model of the basal ganglia can in- CONCLUSION
clude an algorithmic-like RL module to estimate pre-
diction errors. Moreover, the different types of model may In the present review, we have outlined how computa-
beneficially be used in a joint fashion. This approach en- tional neuroscience can use formal models in different
ables us to refine and improve the details of one level of levels to present a mechanistic and functional outlook
characterization limited by the other. For instance, bio- and interpretation on psychopathology and its underlying
physically detailed models of basal ganglia discern be- pathophysiology. We focused on theory-driven appro-
tween opponent indirect and direct pathways that dis- aches and introduced a new category for these computa-
tinctively process dopaminergic reinforcement signals tional models, which require substantial skills from cel-
[74]. The integration of this feature in more abstract algo- lular and molecular neuroscience, network neuroscience,
rithmic models makes it possible to computationally ana- cognitive neuroscience, computational neuroscience,
lyze its consequences for various behaviors across a psychiatry and psychology, computer science, mathe-
broad range of parameters. Furthermore, it also should be matics and engineering to empower the important inter-
noted that Bayesian approaches can be applied to all disciplinary field of computational psychiatry. Ongoing
three subtypes of models for fitting, description, vali- progress in computational neuroscience offers remark-
dation and other purposes. able opportunities to design and conduct mechanistic
Although theory-driven modeling approaches have preclinical studies by focusing on different brain circuits
yielded important insights into the processes and mecha- models, especially in neurodevelopmental disorders such
nisms underlying specific disorders at many levels of anal- as schizophrenia, ADHD and autism. These mechanistic
ysis, they have to be applied yet to clinical problems for studies may be administered at the level of high level
the most part. Furthermore, these computational tools manifestations of mental illness using phenotypic models
have some limitations. Most obvious limitation is that they and neurorobotics platform, at the level of low level brain
require considerable expertise and are often opaque and functioning using neurobiological models, or at the inter-
confusing for the non-expert. Hence, how to exchange mediary level by establishing interrelationships between
useful information between clinicians, trialists, experi- the low level and the high level models using different
mentalists and theorists is challenging for the field. This types of intermediate models. It is our belief that some
34 A. Khaleghi, et al.

challenges in psychiatry and mental health in general adolescents. Iran J Psychiatry 2019;14:1-15.
cannot be overcome without the help of theoretical 6. Mohammadi MR, Ahmadi N, Khaleghi A, Zarafshan H,
Mostafavi SA, Kamali K, et al. Prevalence of autism and its co-
frameworks and computational modeling. However, the
morbidities and the relationship with maternal psychopathol-
practical applications of computational models are still in
ogy: a national population-based study. Arch Iran Med 2019;
their infancy, and there is a long way to go to reach the ul- 22:546-553.
timate goal of guiding the design and development of new 7. Zarafshan H, Khaleghi A, Mohammadi MR, Moeini M, Malmir
pharmacological or even cognitive therapeutic strategies N. Electroencephalogram complexity analysis in children
through the mechanistic interpretability of computational with attention-deficit/hyperactivity disorder during a visual
models. cognitive task. J Clin Exp Neuropsychol 2016;38:361-369.
8. Khaleghi A, Zarafshan H, Mohammadi MR. Visual and audi-
tory steady-state responses in attention-deficit/hyperactivity
■ Funding
disorder. Eur Arch Psychiatry Clin Neurosci 2019;269:645-655.
None. 9. Mostafavi SA, Khaleghi A, Mohammadi MR. Noninvasive
brain stimulation in alcohol craving: a systematic review and
■ Conflicts of Interest meta-analysis. Prog Neuropsychopharmacol Biol Psychiatry
No potential conflict of interest relevant to this article 2020;101:109938.
10. Khaleghi A, Pirzad Jahromi G, Zarafshan H, Mostafavi SA,
was reported.
Mohammadi MR. Effects of transcranial direct current stim-
ulation of prefrontal cortex on risk-taking behavior. Psychiatry
■ Author Contributions Clin Neurosci 2020;74:455-465.
Conceptualization, writing draft, reviewing and editing 11. Khaleghi A, Zarafshan H, Vand SR, Mohammadi MR. Effects
final manuscript: Ali Khaleghi, Mohammad Reza of non-invasive neurostimulation on autism spectrum dis-
Mohammadi, Kian Shahi and Ali Motie Nasrabadi. order: a systematic review. Clin Psychopharmacol Neurosci
2020;18:527-552.
12. Wang XJ, Krystal JH. Computational psychiatry. Neuron
■ ORCID 2014;84:638-654.
Ali Khaleghi [Link] 13. Friston KJ, Stephan KE, Montague R, Dolan RJ. Computational
Mohammad Reza Mohammadi psychiatry: the brain as a phantastic organ. Lancet Psychiatry
[Link] 2014;1:148-158.
Kian Shahi [Link] 14. Huys QJ, Maia TV, Frank MJ. Computational psychiatry as a
bridge from neuroscience to clinical applications. Nat Neurosci
Ali Motie Nasrabadi [Link]
2016;19:404-413.
15. Jaeger D, Jung R. Encyclopedia of computational neuroscience.
REFERENCES New York:Springer;2015.
1. Maia TV, Cano-Colino M. The role of serotonin in orbitofrontal 16. Dayan P, Abbott LF. Book review: Theoretical neuroscience:
function and obsessive-compulsive disorder. Clin Psychol Sci computational and mathematical modeling of neural systems.
2015;3:460-482. J Cognit Neurosci 2003;15:154-155.
2. Stephan KE, Bach DR, Fletcher PC, Flint J, Frank MJ, Friston KJ, 17. Kriegeskorte N, Douglas PK. Cognitive computational neuro-
et al. Charting the landscape of priority problems in psychiatry, science. Nat Neurosci 2018;21:1148-1160.
part 1: classification and diagnosis. Lancet Psychiatry 2016;3: 18. Maia TV, Frank MJ. From reinforcement learning models to
77-83. psychiatric and neurological disorders. Nat Neurosci 2011;
3. Talepasand S, Mohammadi MR, Alavi SS, Khaleghi A, Sajedi 14:154-162.
Z, Akbari P, et al. Psychiatric disorders in children and adoles- 19. Deserno L, Heinz A, Schlagenhauf F. Computational approaches
cents: prevalence and sociodemographic correlates in Semnan to schizophrenia: a perspective on negative symptoms.
Province in Iran. Asian J Psychiatr 2019;40:9-14. Schizophr Res 2017;186:46-54.
4. Khaleghi A, Mohammadi MR, Zandifar A, Ahmadi N, Alavi 20. Valton V, Romaniuk L, Douglas Steele J, Lawrie S, Seriès P.
SS, Ahmadi A, et al. Epidemiology of psychiatric disorders in Comprehensive review: computational modelling of schizo-
children and adolescents; in Tehran, 2017. Asian J Psychiatr phrenia. Neurosci Biobehav Rev 2017;83:631-646.
2018;37:146-153. 21. Yahata N, Kasai K, Kawato M. Computational neuroscience
5. Mohammadi MR, Ahmadi N, Khaleghi A, Mostafavi SA, approach to biomarkers and treatments for mental disorders.
Kamali K, Rahgozar M, et al. Prevalence and correlates of psy- Psychiatry Clin Neurosci 2017;71:215-237.
chiatric disorders in a national survey of Iranian children and 22. Gold A, Dudai Y. Simulation of mental disorders: II. Computer
 Computational Neuroscience Approach to Psychiatry 35

models, purposes and future directions. Isr J Psychiatry Relat 37. Anticevic A, Gancsos M, Murray JD, Repovs G, Driesen NR,
Sci 2016;53:73-81. Ennis DJ, et al. NMDA receptor function in large-scale anti-
23. Karbalaei Akbari M, Zhuiykov S. A bioinspired optoelectroni- correlated neural systems with implications for cognition and
cally engineered artificial neurorobotics device with sensor- schizophrenia. Proc Natl Acad Sci U S A 2012;109:16720-
imotor functionalities. Nat Commun 2019;10:3873. 16725.
24. Falotico E, Vannucci L, Ambrosano A, Albanese U, Ulbrich S, 38. Frank MJ. Dynamic dopamine modulation in the basal gan-
Vasquez Tieck JC, et al. Connecting artificial brains to robots glia: a neurocomputational account of cognitive deficits in
in a comprehensive simulation framework: the neurorobotics medicated and nonmedicated Parkinsonism. J Cogn Neurosci
platform. Front Neurorobot 2017;11:2. 2005;17:51-72.
25. Eliasmith C, Stewart TC, Choo X, Bekolay T, DeWolf T, Tang 39. Gurney KN, Humphries MD, Redgrave P. A new framework
Y, et al. A large-scale model of the functioning brain. Science for cortico-striatal plasticity: behavioural theory meets in vitro
2012;338:1202-1205. data at the reinforcement-action interface. PLoS Biol 2015;13:
26. Zuckerman O, Hoffman G, Kopelman-Rubin D, Klomek AB, e1002034.
Shitrit N, Amsalem Y, et al. KIP3: robotic companion as an ex- 40. Frank MJ. Linking across levels of computation in model-based
ternal cue to students with ADHD. In: TEI '16: Tenth Interna- cognitive neuroscience. In: Forstmann B, Wagenmakers EJ,
tional Conference on Tangible, Embedded, and Embodied editors. An introduction to model-based cognitive neuroscience.
Interaction; Feb 14-17, 2016; Eindhoven, Netherlands. p. New York:Springer;2015. p.159-177.
621-626. 41. Chen C, Takahashi T, Nakagawa S, Inoue T, Kusumi I.
27. Rakhymbayeva N, Seitkazina N, Turabayev D, Pak A, Reinforcement learning in depression: a review of computa-
Sandygulova A. A long-term study of robot-assisted therapy tional research. Neurosci Biobehav Rev 2015;55:247-267.
for children with severe autism and ADHD. In: HRI '20: 42. Deserno L, Boehme R, Heinz A, Schlagenhauf F. Reinforce-
Companion of the 2020 ACM/IEEE International Conference ment learning and dopamine in schizophrenia: dimensions of
on Human-Robot Interaction; March 23-26, 2020, Cambridge, symptoms or specific features of a disease group? Front Psychiatry
UK. p. 401-402. 2013;4:172.
28. O’Reilly RC. Biologically based computational models of 43. Maia TV. Reinforcement learning, conditioning, and the
high-level cognition. Science 2006;314:91-94. brain: successes and challenges. Cogn Affect Behav Neurosci
29. Markram H. The human brain project. Sci Am 2012;306:50-55. 2009;9:343-364.
30. Tsoutsouras V, Sirakoulis GC, Pavlos GP, Iliopoulos AC. 44. Eshel N, Bukwich M, Rao V, Hemmelder V, Tian J, Uchida N.
Simulation of healthy and epileptiform brain activity using Arithmetic and local circuitry underlying dopamine prediction
cellular automata. Int J Bifurc Chaos 2012;22:1250229. errors. Nature 2015;525:243-246.
31. Javanbakht A, Alberini CM. Editorial: Neurobiological models 45. Montague PR, Hyman SE, Cohen JD. Computational roles for
of psychotherapy. Front Behav Neurosci 2019;13:144. dopamine in behavioural control. Nature 2004;431:760-767.
32. Zilcha-Mano S, Roose SP, Brown PJ, Rutherford BR. Not just 46. Moustafa AA, Sherman SJ, Frank MJ. A dopaminergic basis for
nonspecific factors: the roles of alliance and expectancy in working memory, learning and attentional shifting in
treatment, and their neurobiological underpinnings. Front Parkinsonism. Neuropsychologia 2008;46:3144-3156.
Behav Neurosci 2019;12:293. 47. Schultz W, Preuschoff K, Camerer C, Hsu M, Fiorillo CD, Tobler
33. Iglesias S, Tomiello S, Schneebeli M, Stephan KE. Models of PN, et al. Explicit neural signals reflecting reward uncertainty.
neuromodulation for computational psychiatry. Wiley Interdiscip Philos Trans R Soc Lond B Biol Sci 2008;363:3801-3811.
Rev Cogn Sci 2017;8:e1420. 48. Adams RA, Huys QJ, Roiser JP. Computational psychiatry: to-
34. Murray JD, Anticevic A, Gancsos M, Ichinose M, Corlett PR, wards a mathematically informed understanding of mental
Krystal JH, et al. Linking microcircuit dysfunction to cognitive illness. J Neurol Neurosurg Psychiatry 2016;87:53-63.
impairment: effects of disinhibition associated with schizo- 49. Murray GK, Corlett PR, Clark L, Pessiglione M, Blackwell AD,
phrenia in a cortical working memory model. Cereb Cortex Honey G, et al. Substantia nigra/ventral tegmental reward pre-
2014;24:859-872. diction error disruption in psychosis. Mol Psychiatry 2008;13:
35. Lisman JE, Coyle JT, Green RW, Javitt DC, Benes FM, Heckers 239, 267-276.
S, et al. Circuit-based framework for understanding neuro- 50. Corlett PR, Murray GK, Honey GD, Aitken MR, Shanks DR,
transmitter and risk gene interactions in schizophrenia. Trends Robbins TW, et al. Disrupted prediction-error signal in psy-
Neurosci 2008;31:234-242. chosis: evidence for an associative account of delusions.
36. Krystal JH, Karper LP, Seibyl JP, Freeman GK, Delaney R, Brain 2007;130(Pt 9):2387-2400.
Bremner JD, et al. Subanesthetic effects of the noncompetitive 51. Romaniuk L, Honey GD, King JR, Whalley HC, McIntosh AM,
NMDA antagonist, ketamine, in humans. Psychotomimetic, Levita L, et al. Midbrain activation during Pavlovian con-
perceptual, cognitive, and neuroendocrine responses. Arch ditioning and delusional symptoms in schizophrenia. Arch
Gen Psychiatry 1994;51:199-214. Gen Psychiatry 2010;67:1246-1254.
36 A. Khaleghi, et al.

52. Schlagenhauf F, Sterzer P, Schmack K, Ballmaier M, Rapp M, Conscious Cogn 2014;25:67-76.

Wrase J, et al. Reward feedback alterations in unmedicated 68. Mokhtari Hashtjini M, Pirzad Jahromi G, Sadr SS, Khaleghi A,
schizophrenia patients: relevance for delusions. Biol Psychiatry Hatef B, Meftahi GH. Comparison of the effects of deep brain
2009;65:1032-1039. stimulation of the prelimbic cortex and basolateral amygdala
53. Kapur S. Psychosis as a state of aberrant salience: a framework for facilitation of extinction process of conditioned fear. Arch
linking biology, phenomenology, and pharmacology in Neurosci 2020;7.
schizophrenia. Am J Psychiatry 2003;160:13-23. 69. Huys QJM, Dayan P. A Bayesian formulation of behavioral
54. Waltz JA, Schweitzer JB, Gold JM, Kurup PK, Ross TJ, control. Cognition 2009;113:314-328.
Salmeron BJ, et al. Patients with schizophrenia have a reduced 70. Maia TV. Fear conditioning and social groups: statistics, not
neural response to both unpredictable and predictable primary genetics. Cogn Sci 2009;33:1232-1251.
reinforcers. Neuropsychopharmacology 2009;34:1567-1577. 71. Browning M, Behrens TE, Jocham G, O’Reilly JX, Bishop SJ.
55. Gradin VB, Kumar P, Waiter G, Ahearn T, Stickle C, Milders Anxious individuals have difficulty learning the causal sta-
M, et al. Expected value and prediction error abnormalities in tistics of aversive environments. Nat Neurosci 2015;18:590-
depression and schizophrenia. Brain 2011;134(Pt 6):1751- 596.
1764. 72. Fornito A, Zalesky A. Computational approaches to under-
56. Gold JM, Waltz JA, Matveeva TM, Kasanova Z, Strauss GP, standing mental dysfunction: progress, challenges, and new
Herbener ES, et al. Negative symptoms and the failure to rep- frontiers. Biol Psychiatry Cogn Neurosci Neuroimaging 2018;
resent the expected reward value of actions: behavioral and 3:728-730.
computational modeling evidence. Arch Gen Psychiatry 73. Einevoll GT, Destexhe A, Diesmann M, Grün S, Jirsa V, de
2012;69:129-138. Kamps M, et al. The scientific case for brain simulations.
57. Gold JM, Strauss GP, Waltz JA, Robinson BM, Brown JK, Frank Neuron 2019;102:735-744.
MJ. Negative symptoms of schizophrenia are associated with 74. Collins AG, Frank MJ. Opponent actor learning (OpAL): mod-
abnormal effort-cost computations. Biol Psychiatry 2013;74: eling interactive effects of striatal dopamine on reinforcement
130-136. learning and choice incentive. Psychol Rev 2014;121:337-
58. Garbusow M, Schad DJ, Sebold M, Friedel E, Bernhardt N, 366.
Koch SP, et al. Pavlovian-to-instrumental transfer effects in the 75. Mohammadi MR, Khaleghi A, Nasrabadi AM, Rafieivand S,
nucleus accumbens relate to relapse in alcohol dependence. Begol M, Zarafshan H. EEG classification of ADHD and nor-
Addict Biol 2016;21:719-731. mal children using non-linear features and neural network.
59. Sonuga-Barke EJ. The dual pathway model of AD/HD: an Biomed Eng Lett 2016;6:66-73.
elaboration of neuro-developmental characteristics. Neurosci 76. Khaleghi A, Sheikhani A, Mohammadi MR, Nasrabadi AM,
Biobehav Rev 2003;27:593-604. Vand SR, Zarafshan H, et al. EEG classification of adolescents
60. Smith AJ, Becker S, Kapur S. A computational model of the with type I and type II of bipolar disorder. Australas Phys Eng
functional role of the ventral-striatal D2 receptor in the ex- Sci Med 2015;38:551-559.
pression of previously acquired behaviors. Neural Comput 77. Mostafavi SA, Khaleghi A, Mohammadi MR, Akhondzadeh S.
2005;17:361-395. Is transcranial direct current stimulation an effective modality
61. Huys QJ, Pizzagalli DA, Bogdan R, Dayan P. Mapping anhe- in reducing food craving? A systematic review and meta-analysis.
donia onto reinforcement learning: a behavioural meta- Nutr Neurosci 2020;23:55-67.
analysis. Biol Mood Anxiety Disord 2013;3:12. 78. Moeini M, Khaleghi A, Mohammadi MR. Characteristics of al-
62. Yoshida W, Dziobek I, Kliemann D, Heekeren HR, Friston KJ, pha band frequency in adolescents with bipolar II disorder: a
Dolan RJ. Cooperation and heterogeneity of the autistic mind. resting-state QEEG study. Iran J Psychiatry 2015;10:8-12.
J Neurosci 2010;30:8815-8818. 79. Khaleghi A, Birgani PM, Fooladi MF, Mohammadi MR. Applicable
63. Kishida KT, Montague PR. Imaging models of valuation during features of electroencephalogram for ADHD diagnosis. Res
social interaction in humans. Biol Psychiatry 2012;72:93-100. Biomed Eng 2020;36:1-11.
64. Kishida KT, King-Casas B, Montague PR. Neuroeconomic ap- 80. Mohammadi MR, Khaleghi A. Transsexualism: a different
proaches to mental disorders. Neuron 2010;67:543-554. viewpoint to brain changes. Clin Psychopharmacol Neurosci
65. Sanfey AG. Social decision-making: insights from game theo- 2018;16:136-143.
ry and neuroscience. Science 2007;318:598-602. 81. Wiecki TV, Antoniades CA, Stevenson A, Kennard C, Borowsky
66. Yoshida W, Seymour B, Friston KJ, Dolan RJ. Neural mecha- B, Owen G, et al. Computational cognitive biomarker for ear-
nisms of belief inference during cooperative games. J Neurosci ly-stage Huntington’s disease. PLoS One 2016;11:e0148409.
2010;30:10744-10751. 82. Brodersen KH, Deserno L, Schlagenhauf F, Lin Z, Penny WD,
67. Moutoussis M, Fearon P, El-Deredy W, Dolan RJ, Friston KJ. Buhmann JM, et al. Dissecting psychiatric spectrum disorders
Bayesian inferences about the self (and others): a review. by generative embedding. Neuroimage Clin 2013;4:98-111.