Ligand bias of scoring functions in structure-based virtual screening

Area of Science:

• Computational chemistry
• Drug discovery

Background:

• Structure-based virtual screening (SBVS) relies on scoring functions to predict ligand-target binding.
• Scoring functions can exhibit ligand bias, leading to false positives and reduced screening efficiency.

Purpose of the Study:

• To evaluate postprocessing methods for improving enrichment factors in virtual screening.
• To investigate the effectiveness of multiple active site correction (MASC) and partial least squares (PLS) MASC in mitigating ligand bias.

Main Methods:

• Docking of known actives and decoy compounds to eight targets using 10 scoring functions.
• Application of multiple active site correction (MASC), PLS MASC, and size normalization techniques.
• Analysis of enrichment factors and correlation between scores and molecular descriptors.

Main Results:

• Standard MASC did not consistently improve enrichment.
• Unit variance normalization and PLS MASC showed significant performance improvements.
• PLS MASC was more effective when scoring functions exhibited size dependence.
• Normalization by molecular weight or heavy atom count, especially with root transformations, enhanced enrichment.

Conclusions:

• Ligand bias in scoring functions is a significant source of false positives in SBVS.
• PLS MASC and specific normalization strategies can effectively reduce false positives and improve virtual screening performance.