CaliPPer Framework for Binding Prediction Performance Estimation

CaliPPer provides post-hoc performance estimation for binding prediction models using Sample-to-Domain Distance and Bayesian recalibration. It achieves distance-performance correlations of |r|=0.80–0.92 and predicts AUROC/AP/F1 metrics with low mean absolute error across ten architectures.