Sequence-based drug-target binding site pre-training enables cryptic pocket detection and improves binding affinity...

Area of Science:

• Computational chemistry
• Structural biology
• Drug discovery

Background:

• Accurate characterization of protein-ligand interactions is crucial for drug discovery.
• Existing computational methods have limitations in data integration, protein dynamics representation, and interaction modeling.

Purpose of the Study:

• Introduce ProMoNet, a sequence-based framework to improve predictions of protein-ligand binding characteristics.
• Address limitations in current computational drug discovery tools.

Main Methods:

• Developed ProMoNet, a sequence-based pre-training and fine-tuning framework.
• Utilized protein and molecular foundation models for expanded data coverage.
• Implemented a pre-training strategy based on protein-ligand binding site prediction.
• Designed pre-training and fine-tuning modules to model microscale interactions and protein dynamics.

Main Results:

• ProMoNet effectively models microscale protein-ligand interactions and protein dynamics, including binding site crypticity, without 3D structures.
• The pre-training module matches or surpasses state-of-the-art structure-based methods in binding site identification.
• The fine-tuning module achieves superior performance in binding affinity and kinetics prediction.

Conclusions:

• ProMoNet demonstrates strong performance and efficiency across multiple protein-ligand binding prediction tasks.
• The framework shows significant potential for accelerating drug discovery applications.
• ProMoNet offers a promising computational tool for predicting binding characteristics.