Design of Synthesizable PROTACs through Synthesis Constrained Generative Model and Reinforcement Learning

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Drug Discovery

Background:

• Proteolysis Targeting Chimeras (PROTACs) offer a novel therapeutic strategy by degrading disease-related proteins.
• Current deep generative models for PROTAC design often yield molecules with poor synthetic feasibility.
• Addressing this gap is crucial for advancing PROTAC technology in drug development.

Purpose of the Study:

• To develop a novel computational model, SynPROTAC, for designing synthesizable PROTACs with desirable binding properties.
• To integrate chemical reaction path-driven assembly with reinforcement learning for efficient PROTAC generation.
• To validate the model's capability in designing and synthesizing novel bioactive PROTACs.

Main Methods:

• SynPROTAC employs a synthesis-constrained generative approach using a Graph Transformer-encoded input (warhead or E3 ligand).
• The model autoregressively samples reaction templates and building blocks via a transformer-based decoder for PROTAC construction.
• Reinforcement learning is integrated to optimize for synthesizability and binding properties.

Main Results:

• SynPROTAC successfully generated novel PROTACs with feasible synthetic routes and favorable physicochemical and binding properties.
• Two designed PROTAC molecules targeting bromodomain-containing protein 4 (BRD4) were synthesized following the model's proposed routes.
• Both synthesized compounds demonstrated nanomolar-level BRD4 degradation and potent antiproliferation activity against MV411 tumor cells.

Conclusions:

• SynPROTAC effectively designs novel, bioactive PROTAC molecules with practical synthetic pathways.
• The model represents a significant advancement in accelerating the development of synthesizable PROTAC therapeutics.
• This work demonstrates the potential of AI-driven approaches for efficient drug discovery and design.