A New Amino Acid Similarity Matrix Based on Sequence Correlations and Structural Features Yields Complete Sequence-Structure Congruence

Abstract

Protein sequence matching does not properly account for some well-known features of protein structures: surface residues being more variable than core residues, the high packing densities in globular proteins, and does not yield good matches of sequences of many proteins known to be close structural relatives. There are now abundant protein sequences and structures to enable major improvements to sequence matching. Here, we utilize structural frameworks to mount the observed correlated sequences to identify the most important correlated parts. The rationale is that protein structures provide the important physical framework for improving sequence matching. Combining the sequence and structure data in this way enables the incorporation of allosteric information into sequence matching and transforms it effectively from a 1-D to a 3-D procedure. Our results show that there are major gains in the specificity of sequence matching across diverse proteins. Specifically, all known cases where protein structures match but sequences do not match well are resolved.