THRUST 3

The overarching goal of Thrust 3 is to pressure test and advance the foundational AI agents developed in Thrust 1 by applying them to drug discovery and synthesis. Drug discovery represents an ideal testbed because it requires simultaneous reasoning over chemical function, synthetic feasibility, and emergent molecular properties.

In this thrust, we will deploy the modular Chemical Language Model (mCLM) to discover new kinase inhibitors (KIs) and validate that the mCLM can reason over function-infused molecular modules that are compatible with automated modular synthesis. This approach complements other CLMs that do not account for synthesis compatibility a priori1–10 and other modular discovery strategies such as DNA-encoded libraries (which do not consider function a priori) and fragment-based drug discovery (which does not consider iterative automated synthesis a priori.

Additionally, we will use AI-enabled synthesis planning tools to design and experimentally validate highly efficient chemoenzymatic routes to three FDA-approved drugs. Finally, we will develop new LLM-enabled models to generate robust retrosynthetic pathways, experimental procedures, and predictions of enzyme selectivity.