Construction of a virtual nigh throughput screen by 4D-QSAR analysis: Application to a combinatorial library of glucose inhibitors of glycogen phosphorylase b

The 4D-QSAR model developed for a training set of 47 glucose analogue inhib itors of glycogen phosphorylase, and reported in the previous paper in this issue, was used as a basis for developing virtual high throughput screen, VHTS, models to screen a focused combinatorial virtual library of 225 addit ional inhibitors. Techniques to develop, evaluate, and apply VHTS models de rived from 4D-QSAR models are presented. Application of the VHTS models to screen the virtual library results in the prediction of compounds which bin d both more, and less, strongly to the enzyme than the best and worst binde rs of the training set. Analysis of the binding predictions across the virt ual library reveals patterns of structure-activity information that can be useful to design new focused libraries. The possible use of overfit QSAR mo dels, with respect to the training data set, as VHTS models is discussed an d explored.