Template CoMFA Applied to 116 Biological Targets

Statistically
acceptable 3D-QSAR models were effortlessly obtained,
by application of automatic template CoMFA, for the vast majority,
arguably more than 95%, of 116 biological targets. Among these targets,
76 were structure-based, pooling multiple templates (as protein-bound
conformation of ligands) and training sets into a single model, and
the other 40 were ligand-based, with a low-energy conformation of
an exemplar training set structure becoming the single template. Criteria
proposed for statistical acceptability are a leave-one-out q2 > 0.4, or a standard error of leave-one-out
prediction <1.0, or a ratio of best-fit r2 to count of PLS components >0.2. The structure-based 3D-QSAR
models provide direct visual comparisons of SAR-derived 3D-QSAR contours
with cavity surfaces.