Rescoring and Linearly Combining: A Highly Effective Consensus Strategy for Virtual Screening Campaigns

Area of Science:

• Computational chemistry
• Cheminformatics
• Drug discovery

Background:

• Virtual screening is crucial for identifying drug candidates.
• Existing docking scoring functions have limitations in accuracy.
• Improving the reliability of virtual screening is a key challenge.

Purpose of the Study:

• To develop and evaluate a novel consensus strategy for virtual screening.
• To enhance the performance of docking score evaluations using linear combinations.
• To optimize consensus models using the Enrichment Factor Optimization (EFO) method.

Main Methods:

• Generated consensus models by linearly combining multiple docking scores.
• Utilized the Enrichment Factor Optimization (EFO) method to develop linear equations.
• Evaluated performance using the Directory of Useful Decoys (DUD) datasets.
• Assessed model reliability via per-target validation.

Main Results:

• Consensus strategy significantly improved Top 1% enrichment factors (average 17% increase).
• Kinase targets showed substantial improvement, with EF increasing from 6.4 to 23.5.
• Area Under the Curve (AUC) improved by 14% on average when combining three scores.
• Three-variable models achieved the highest AUC, comparable to recent literature benchmarks.

Conclusions:

• The proposed consensus strategy effectively enhances virtual screening performance.
• Linear combinations of docking scores offer a reliable approach to improve hit identification.
• This method provides a robust tool for computational drug discovery.