Improved identification of cancer driver mutations using evolutionary patterns of local functional spatial regions on protein structures.

Abstract

Protein sequences provide rich information for understanding protein functions. An important task is to quantitatively characterize the evolutionary patterns that is directly under the selection pressure biased towards biochemical functions. Here we employ a continuous time Markov model to describe the evolutionary process of protein, and estimate full sequence length and region-specific substitution rates of residues on these surface pockets. Our method is validated by accurate reconstruction of the Jones-Taylor-Thorton Matrix. In addition, our method can accurately identify alpha-amylase that all perform similar biochemical functions through evolutionary analysis of surface pockets on their PDB structure. Furthermore, we show that there are characteristic patterns of evolution on local regions of 242 confirmed cancer driver mutation genes. We also discuss how these evolutionary patterns compare with those derived from full protein sequences. In addition, we show how new scoring matrix based on these local regional evolutionary patterns can improve sensitivity of prediction of cancer driver mutations.