3D based generative PROTAC linker design with reinforcement learning.

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Drug Discovery

Background:

• Proteolysis targeting chimeras (PROTACs) leverage the ubiquitin-proteasome system for selective protein degradation.
• PROTACs feature a warhead for the protein of interest (POI) and an E3-ligase ligand, connected by a linker.
• Rational linker design is complex, often lacking 3D structural information.

Purpose of the Study:

• To introduce PROTAC-INVENT, a novel 3D generative model for PROTAC linker design.
• To generate PROTAC structures with putative 3D binding conformations.
• To enhance the efficiency of PROTAC development through advanced computational methods.

Main Methods:

• Development of PROTAC-INVENT, a 3D generative model for PROTACs.
• Joint training with reinforcement learning (RL) to optimize 2D and 3D properties.
• Generation of SMILES and 3D conformations for PROTACs, target proteins, and E3 ligases.

Main Results:

• PROTAC-INVENT successfully generates PROTAC SMILES and corresponding 3D binding conformations.
• The model demonstrates utility in creating reasonable 3D PROTAC structures.
• The 3D conformation generation workflow serves as an efficient PROTAC docking protocol.

Conclusions:

• PROTAC-INVENT offers a novel approach to rational PROTAC linker design.
• The model facilitates the generation of 3D PROTAC conformations, aiding drug discovery.
• The developed workflow enhances PROTAC docking and design efficiency.