Prediction of Water's Mobility and Disorder in Protein Crystals Using Novel Local Hydrophobic Descriptors.

Abstract

The B-factors of crystal structures reflect the atomic fluctuations about their average positions and provide important information about molecular dynamics. Although numerous works have been addressed on theoretical and computational studies of B-factor profile of protein atoms, the methods used for predicting B-factor values of water molecules in protein crystals still remain unexploited. In this article, we describe a new approach that we named local hydrophobic descriptors (LHDs) to characterize the hydrophobic landscapes of protein hydration sites. Using this approach coupled with partial least squares (PLS) regression and least-squares squares support vector machine (LSSVM), we perform a systematic investigation on the linear and nonlinear relationships between the LHDs and water B-factors. Based upon an elaborately selected, large-scale dataset of crystal water molecules, our method predicts B-factor profile with coefficient of determination rpred of 0.554. We demonstrate that (i) the dynamics of water molecules is primarily governed by the local features of hydrophobic potential landscapes, and (ii) the accuracy of predicted B-factor values depends on water packing density.