Robust Classification of “Relevant” Water Molecules in Putative Protein Binding Sites

Abstract

A statistically validated protocol to identify "relevant" water molecules in protein binding sites using HINT score and a geometric descriptor termed Rank is described. In training, conservation/nonconservation was modeled for 86% of the waters. For the test set, 87% of waters were correctly classified (92% when crystallographic resolution was <or=2.0 A). Conserved waters make at least two hydrogen bonds with protein and gain 0.6-2.0 kcal mol(-1) more binding energy than nonconserved waters.