Tools to build molecular-docking activity prediction models by PLS regression with iterative training and pose-selection. Descriptors include (i) docking score(s), (ii) pharmacophore features, (iii) multi-feature descriptors learned by decision trees.