Improved protein-ligand binding affinity prediction by using a curvature-dependent surface-area model

Motivation: Hydrophobic effect plays a pivotal role in most protein-ligand binding. State-of-the-art protein-ligand scoring methods usually treat hydrophobic free energy as surface tension, which is proportional to interfacial surface area for simplicity and efficiency. However, this treatment ignores the role of molecular shape, which has been found very important by either experimental or theoretical studies. Results: We propose a new empirical scoring function, named Cyscore. Cyscore improves the prediction accuracy by using a novel curvature-dependent surface-area model, which is able to distinguish convex, planar and concave surface in hydrophobic free energy calculation. Benchmark tests show that this model significantly improves the protein-ligand scoring and Cyscore outperforms a variety of well established scoring functions using PDBbind benchmark sets for binding affinity correlation and ranking tests. We expect the curvature-dependent surface-area model and Cyscore would contribute to the study of protein-ligand interactions.