PREDICTION OF PROTEIN HYDRATION SITES FROM SEQUENCE BY MODULAR NEURALNETWORKS

The hydration properties of a protein are important determinants of it s structure and function. Here, modular neural networks are employed t o predict ordered hydration sites using protein sequence information. First, secondary structure and solvent accessibility are predicted fro m sequence with two separate neural networks. These predictions are us ed as input together with protein sequences for networks predicting hy dration of residues, backbone atoms and sidechains. These networks are trained with protein crystal structures. The prediction of hydration is improved by adding information on secondary structure and solvent a ccessibility and, using actual values of these properties, residue hyd ration can be predicted to 77% accuracy with a Matthews coefficient of 0.43. However, predicted property data with an accuracy of 60-70% res ult in less than half the improvement in predictive performance observ ed using the actual values. The inclusion of property information allo ws a smaller sequence window to be used in the networks to predict hyd ration. It has a greater impact on the accuracy of hydration site pred iction for backbone atoms than far sidechains and for non-polar than p olar residues. The networks provide insight into the mutual interdepen dencies between the location of ordered water sites and the structural and chemical characteristics of the protein residues.