Overall horrible enrichment for an original dud target (mr_auto) that performed well. However, old ligands and decoys docked to the new builds perform well (~40 LogAUC), so the new ligands and decoys are just much more challenging. Since enrichment on the new dataset is neglible in all cases, and a lot of the structures are mutants anyways, I choose 2aa2 which is the same as pdb code as the old DUD, and not surprisingly enriches the older DUD ligands and decoys well.