Druggability indices for protein targets derived from NMR-based screening data.

An analysis of heteronuclear-NMR-based screening data is used to derive relationships between the ability of small molecules to bind to a protein and various parameters that describe the protein binding site. It is found that a simple model including terms for polar and apolar surface area, surface complexity, and pocket dimensions accurately predicts the experimental screening hit rates with an R(2) of 0.72, an adjusted R(2) of 0.65, and a leave-one-out Q(2) of 0.56. Application of the model to predict the druggability of protein targets not used in the training set correctly classified 94% of the proteins for which high-affinity, noncovalent, druglike leads have been reported. In addition to understanding the pocket characteristics that contribute to high-affinity binding, the relationships that have been defined allow for quantitative comparative analyses of protein binding sites for use in target assessment and validation, virtual ligand screening, and structure-based drug design.