SFCscore(RF): a random forest-based scoring function for improved affinity prediction of protein-ligand complexes.

Area of Science:

• Computational Chemistry
• Structural Biology
• Drug Discovery

Background:

• Empirical scoring functions often show poor correlation between predicted and experimental protein-ligand binding affinities.
• Improving scoring functions requires novel descriptors, larger high-quality training sets, and advanced regression techniques.

Purpose of the Study:

• To develop an improved empirical scoring function for protein-ligand binding affinity prediction.
• To evaluate the performance of the new scoring function against established benchmarks.

Main Methods:

• Utilized SFCscore descriptors and a PDBbind training set of 1005 complexes.
• Employed random forest for regression to develop the SFCscore(RF) scoring function.
• Performed leave-cluster-out cross-validation and assessed performance on PDBbind and CSAR-NRC HiQ benchmarks.

Main Results:

• The developed SFCscore(RF) demonstrated significantly improved performance compared to previous SFCscore functions.
• Validation on PDBbind and CSAR-NRC HiQ benchmarks confirmed the enhanced predictive accuracy.
• Cross-validation and CSAR 2012 exercise highlighted areas for future refinement.

Conclusions:

• SFCscore(RF) represents a significant advancement in empirical scoring functions for protein-ligand binding affinity prediction.
• The study provides insights into limitations and future directions for improving computational scoring functions in drug discovery.