Combined application of cheminformatics- and physical force field-based scoring functions improves binding affinity...

Area of Science:

• Computational chemistry
• Structural biology
• Drug discovery

Background:

• Accurate prediction of ligand-protein binding affinity is crucial for drug discovery.
• The CSAR-NRC benchmark datasets offer a valuable resource for evaluating scoring functions.
• Existing scoring functions have limitations in predicting binding affinities.

Purpose of the Study:

• To assess the performance of Quantitative Structure Binding Affinity Relationships (QSBAR) models and MedusaScore in predicting ligand binding affinity.
• To investigate the efficacy of combining these scoring functions into a consensus approach.
• To identify factors contributing to prediction inaccuracies.

Main Methods:

• Application of QSBAR models, utilizing protein-ligand interface descriptors, to predict binding affinity.
• Utilizing MedusaScore, a physics-based scoring function, for binding affinity prediction.
• Developing and evaluating a consensus scoring function combining QSBAR and MedusaScore.

Main Results:

• Both QSBAR models and MedusaScore achieved statistically significant prediction accuracies (R(2) of 0.44/0.53 and 0.34/0.47, respectively).
• The consensus scoring function demonstrated superior performance, yielding higher R(2) values (0.45/0.58).
• Specific chemical features and noncovalent interactions were identified as potential sources of prediction errors.

Conclusions:

• Consensus scoring functions integrating QSBAR models and MedusaScore enhance prediction accuracy for ligand binding affinity.
• The study highlights the importance of considering multiple descriptors and physical interactions for robust binding affinity prediction.
• Further investigation into specific ligand-protein interactions is warranted to refine scoring function performance.