Prediction of protein-ligand binding affinities using multiple instance learning.

Area of Science:

• Computational chemistry
• Structural biology
• Drug discovery

Background:

• Accurate prediction of protein-ligand binding affinity is crucial for lead optimization in drug discovery.
• Existing scoring functions for molecular docking simulations face significant challenges in predictive accuracy.
• Developing robust scoring functions is essential for efficient structure-based drug design.

Purpose of the Study:

• To propose a data-driven approach integrating multiple scoring functions for direct prediction of protein-ligand binding affinity.
• To introduce Multiple Instance Regression based Scoring (MIRS), a novel method incorporating unbound ligand conformations.
• To evaluate the predictive performance of MIRS against conventional scoring functions for both binding affinity and pose prediction.

Main Methods:

• Developed a data-driven approach integrating multiple scoring functions.
• Proposed the Multiple Instance Regression based Scoring (MIRS) method.
• Evaluated MIRS using 100 protein-ligand complexes and their experimentally determined binding affinities.

Main Results:

• MIRS demonstrated superior performance compared to 11 conventional scoring functions (LigScore, PLP, AutoDock, G-Score, D-Score, LUDI, F-Score, ChemScore, X-Score, PMF, DrugScore).
• MIRS showed strong performance in predicting binding poses.
• The study confirmed the necessity of including unbound ligand conformations for accurate prediction.

Conclusions:

• Incorporating unbound ligand conformations is indispensable for both binding affinity and binding pose prediction.
• The MIRS method accelerates lead optimization in structure-based drug design.
• This work offers a new direction for the development of advanced scoring functions.