Prediction of hydrophilic and hydrophobic hydration structure of protein by neural network optimized using experimental data

Abstract

The hydration structures of proteins, which are necessary for their folding, stability, and functions, were visualized using X-ray and neutron crystallography and transmission electron microscopy. However, complete visualization of hydration structures over the entire protein surface remains difficult. To compensate for this incompleteness, we developed a three-dimensional convolutional neural network to predict the probability distribution of hydration water molecules on the hydrophilic and hydrophobic surfaces, and in the cavities of proteins. The neural network was optimized using the distribution patterns of protein atoms around the hydration water molecules identified in the high-resolution X-ray crystal structures. We examined the feasibility of the neural network using water sites in the protein crystal structures that were not included in the datasets. The predicted distribution covered most of the experimentally identified hydration sites, with local maxima appearing in their vicinity. This computational approach will help to highlight the relevance of hydration structures to the biological functions and dynamics of proteins.