CWRA

Computational drug discovery relies on virtual screening to efficiently identify HIT compounds for experimental testing. Modern virtual screening campaigns often combine heterogeneous computational methods to predict ligand binding, but optimally integrating these diverse predictions remains challenging. We present a data-driven framework that learns modality-specific weights from known active binders and aggregates ranks using weighted robust rank aggregation with quantified uncertainty. Applied to the Vitamin D Receptor, we fuse eleven computational modalities, including GraphDTA endpoints ($K_d$, $K_i$, IC$_{50}$), MLT-LE $pK_d$, AutoDock Vina, Boltz-2 affinity and confidence, Uni-Mol similarity, and three additional ML affinity predictors (TankBind, DrugBAN, and MolTrans).