Optimal Molecular Design: Generative Active Learning Combining REINVENT with Precise Binding Free Energy Ranking...

Area of Science:

• Computational chemistry and cheminformatics
• Artificial intelligence in drug discovery
• Molecular modeling and simulation

Background:

• Active learning (AL) accelerates molecular discovery by intelligently selecting data points, mimicking the iterative design-make-test-analysis cycle.
• Traditional drug discovery is often lengthy and resource-intensive, necessitating more efficient computational approaches.
• Generative AI and physics-based simulations offer complementary strengths for exploring vast chemical spaces.

Purpose of the Study:

• To develop and deploy a generative active learning (GAL) protocol for discovering novel molecular ligands.
• To combine generative molecular AI (REINVENT) with physics-based simulations (ESMACS) for enhanced ligand discovery.
• To assess the efficiency and effectiveness of the GAL protocol on large-scale computing infrastructure (Frontier).

Main Methods:

• Implementation of a GAL protocol integrating REINVENT for molecular generation and ESMACS for binding free energy calculations.
• Application of the GAL protocol to identify ligands for two target proteins: 3CLpro and TNKS2.
• Systematic variation of batch sizes for free energy assessment to optimize protocol efficiency.

Main Results:

• The GAL protocol successfully identified higher-scoring molecules compared to baseline methods for both 3CLpro and TNKS2 targets.
• Discovered ligands exhibited significant chemical diversity and occupied distinct chemical spaces.
• Analysis provided optimal batch size recommendations for different scenarios within the GAL protocol.

Conclusions:

• The combined AI and physics-based approach offers a powerful strategy for effective chemical space sampling in drug discovery.
• The GAL protocol demonstrates unprecedented scale and efficiency, relevant to modern data-driven drug discovery.
• This methodology accelerates the identification of optimized compounds for specific design tasks.