Binding affinity prediction for protein-ligand complexes based on β contacts and B factor.

Area of Science:

• Biochemistry
• Computational Chemistry
• Structural Biology

Background:

• Accurate prediction of protein-ligand binding affinity is crucial for drug design and molecular docking.
• Existing scoring functions often struggle with prediction accuracy, particularly in leave-one-cluster-out cross-validation (LCOCV).

Purpose of the Study:

• To introduce B2BScore, a novel scoring function designed to enhance protein-ligand binding affinity prediction.
• To improve prediction performance using physicochemical properties like β contacts and B factor.

Main Methods:

• Developed B2BScore integrating β contacts (direct atomic contact area) and B factor (atomic mobility).
• Evaluated B2BScore on the PDBBind2009 dataset using independent testing and LCOCV.
• Utilized random forest learning to identify key contact descriptors.

Main Results:

• B2BScore demonstrated superior prediction performance compared to existing methods on independent data and LCOCV.
• Achieved a significant LCOCV improvement, increasing averaged Pearson's correlation coefficients from 0.418 to 0.518.
• Reduced the standard deviation of coefficients in LCOCV from 0.352 to 0.196, indicating greater robustness.

Conclusions:

• B2BScore offers improved accuracy and robustness for predicting protein-ligand binding affinity.
• Identified key binding descriptors, including specific atom contacts and metal ion interactions, valuable for guiding docking.
• The method shows promise for advancing drug discovery and biochemical research.