Using Mutual Information and Models of Evolution for Improved Pattern Detection

Abstract

Many uses of Information Theory have recently been discovered in the field of bioinformatics -clustering and classification of data, sequence alignment scoring, discovering dependencies between sites in amino acid alignments, etc. Mutual Information has proven itself to be a very convenient metric for determining the dependency between two sets of data, and has advantages over other common statistical methods such as correlation. Models of evolution, or substitution matrices, have always been at the very heart of bioinformatics, with a large variety of applications based on PAM, BLOSUM, JTT or other matrices. In this paper we describe a novel algorithm that incorporates substitution rates from a given matrix when calculating the mutual information between sites in an amino acid alignment. We formally describe this algorithm in detail as well as some experimental results. As a result of this work we demonstrate that the incorporation of substitution matrices in the calculation leads to an improved detection of patterns of similarity between sites within a multiple sequence alignment.