Improving Accuracy of Functional Surfaces of Protein Structures Modeled From Sequences

Abstract

Accurate models of protein functional surfaces are important for understanding enzyme mechanism, protein function prediction, compound-protein docking, and drug design. As the speed of rapid accumulation of protein sequence information far exceeds that of structures, it is important to construct models of proteins with accurate functional surfaces. A promising approach is to construct comparative models. We built three-dimensional comparative models of proteins using MODELLER and applied the pocket algorithm CASTp based on alpha shapes to compute functional surfaces for the modeled protein structures. To improve the accuracy of modeled functional surfaces, we have developed a new side-chain repacking algorithm based on an approximation algorithm that solves the optimization problem of finding compatible side chain rotamers with lowest energy, which is formulated as an integer programming problem. Our result showed that for 9 challenging modeled proteins (backbone sequence identity less than 40% and pocket fragments sequence identity less than 50%) with functional surface atoms between 60∼80, 5 have improved accuracy of functional surfaces after side-chain repacking, with the overall recall increased from 0.193 to 0.203 for all 9 structures.