Calibrated Weighted Rank Aggregation for Virtual Screening Independently Rediscovers Privileged Vitamin D Receptor...

Area of Science:

• Computational drug discovery
• Bioinformatics
• cheminformatics

Background:

• Virtual screening is crucial for prioritizing compounds in drug discovery.
• Combining diverse computational scoring methods for robust ranking remains a challenge.

Purpose of the Study:

• To introduce Calibrated Weighted Rank Aggregation (CWRA), a novel framework for data-driven score fusion.
• To optimize the aggregation of heterogeneous computational drug discovery scores for improved compound prioritization.

Main Methods:

• CWRA employs direction correction and per-modality normalization of scoring outputs.
• Nonnegative fusion weights are learned by optimizing a rank-based objective (BEDROC) under simplex constraints.
• The framework was applied to Vitamin D Receptor (VDR) and GABA A receptor targets, fusing machine learning and structure-based scores.

Main Results:

• CWRA significantly improved early enrichment for VDR actives (EF@1%=21.32±6.41), outperforming single methods and equal-weight fusion.
• Applied to GABA A receptor, CWRA also enhanced early enrichment (EF@1%=25.41±3.96).
• Target-specific calibration within CWRA proved more effective than cross-target transfer.

Conclusions:

• CWRA is an effective target-adaptive framework for improving virtual screening performance.
• The method enhances early prioritization and moderate-depth screening by optimizing score fusion.
• CWRA preferentially selects chemically coherent candidates, recapitulating known pharmacophore patterns.