Conditional molecular dynamics refinement for protein-ligand affinity prediction

Area of Science:

• Computational chemistry
• Structural biology
• Drug discovery

Background:

• Protein-ligand affinity prediction is crucial for drug discovery.
• Existing methods struggle with static conformations, posing challenges for accuracy.
• Pose uncertainty and conformational flexibility limit current prediction models.

Purpose of the Study:

• To develop a dynamics-aware framework for robust protein-ligand affinity prediction.
• To address limitations of static models in capturing conformational dynamics.
• To enhance virtual screening and lead discovery through improved affinity prediction.

Main Methods:

• Pocket-conditioned molecular dynamics refinement to model ligand conformational evolution.
• A dual-view atomic representation capturing intra-molecular and inter-molecular geometries.
• Incorporation of global ligand and pocket representations for comprehensive modeling.

Main Results:

• CMD-PLA demonstrates robust performance across diverse settings.
• The framework explicitly models ligand dynamics as a pocket-dependent process.
• Case studies highlight the model's interpretability.

Conclusions:

• CMD-PLA offers a significant advancement in protein-ligand affinity prediction.
• The dynamics-aware approach enhances accuracy and reliability in virtual screening.
• Explicit modeling of conformational changes improves lead discovery pipelines.