Accelerating drug target inhibitor discovery with a deep generative foundation model.

Area of Science:

• Computational chemistry
• Drug discovery
• Structural biology

Background:

• Discovering inhibitors for novel drug targets is difficult, particularly without structural information or known active molecules.
• Existing methods often require extensive target-specific data, limiting rapid response to emerging threats.

Purpose of the Study:

• To validate a deep generative framework for unbiased, sequence-based small-molecule inhibitor design.
• To assess the framework's efficacy against SARS-CoV-2 targets, specifically the spike protein receptor-binding domain (RBD) and main protease.

Main Methods:

• A large-scale deep generative foundation model was trained on protein sequences, small molecules, and their interactions.
• Sequence-conditioned sampling was employed to design candidate inhibitors based solely on target protein sequences.
• Synthesized compounds were experimentally tested in vitro for inhibitory activity against the target proteins.

Main Results:

• Two out of four synthesized small-molecule candidates showed micromolar-level inhibition for each target.
• The most effective spike protein RBD inhibitor demonstrated activity against multiple SARS-CoV-2 variants in live virus neutralization assays.
• The generative framework successfully identified inhibitors without prior knowledge of target structure or existing binders.

Conclusions:

• A single, broadly applicable deep generative foundation model can efficiently accelerate inhibitor discovery.
• This sequence-based approach is effective even when target structure or active molecule information is unavailable.
• The framework offers a promising strategy for rapid development of therapeutics against emerging infectious diseases.