Apo2Mol: 3D Molecule Generation via Dynamic Pocket-Aware Diffusion Models.

Area of Science:

• Computational chemistry and structural biology
• Artificial intelligence in drug discovery

Background:

• Structure-based drug design (SBDD) utilizes protein structures to develop small molecule ligands.
• Existing generative models often overlook protein binding pocket flexibility, limiting drug design accuracy.
• Protein conformational changes upon ligand binding are crucial for effective drug development.

Purpose of the Study:

• To introduce Apo2Mol, a novel diffusion-based generative framework for 3D molecule design.
• To explicitly incorporate protein binding pocket conformational flexibility into the ligand generation process.
• To enable the simultaneous generation of ligands and their corresponding protein pocket conformations.

Main Methods:

• Curated a dataset of over 24,000 apo-holo protein-ligand structure pairs from the Protein Data Bank.
• Developed a full-atom hierarchical graph-based diffusion model.
• The model generates 3D molecules and predicts holo pocket conformations from apo states.

Main Results:

• Apo2Mol achieves state-of-the-art performance in generating high-affinity ligands.
• The framework accurately captures realistic protein pocket conformational changes induced by ligand binding.
• Demonstrated the model's capability in de novo ligand and pocket conformation generation.

Conclusions:

• Apo2Mol advances generative models for drug design by addressing protein flexibility.
• The approach provides a more realistic simulation of ligand-protein interactions.
• This framework holds significant promise for improving the efficiency and success of practical drug discovery.