Amino acid similarity matrix for homology modeling derived from structural alignment and optimized by the Monte Carlo method

Abstract

In this paper, we obtained a similarity matrix for homology modeling based on the structure of proteins in a structural alignment. The alignment procedure was executed within dynamic programming generally used in alignment methods. An initial matrix derived from the structural alignment was optimized by the Markov chain Monte Carlo method at low temperature to fit its sequence alignment to the structural alignment. Structural alignment was performed on the basis of the superposition of Cα atoms for two protein structures. The objective function in the Monte Carlo procedure was defined by entropy in the information theory, allowing us to show that the amino acid similarity matrix aligned accurately. When compared with the structural alignment, the average number of incorrect amino acid residues in the sequence alignment was 22.6 for all residues and about 3.7 for residues in structurally conserved regions. The alignment with our matrix was more similar to structural alignment than to sequence alignments using other amino acid substitution matrices.