Combining Bayesian and Evidential Uncertainty Quantification for Improved Bioactivity Modeling.

For each bioactivity end point ( xC 50 and K x ) and each split type (stratified and scaffold cluster), we conducted tests separately across models and across the selected performance and uncertainty metrics (RMSE, miscalibration area, NLL, CRPS, interval, sharpness). This dual-testing strategy ensured that conclusions did not depend on a single assumption about distributional form, and aligns with recent recommendations for practically significant model comparisons in molecular machine learning.

also in 1,558 other papers