An entropy-based technique for classifying bacterial chromosomes according to synonymous codon usage

Abstract

We present a framework based on information theoretic concepts and the Dirichlet distribution for classifying chromosomes based on the degree to which they use synonymous codons uniformly or preferentially, that is, whether or not codons that code for an amino acid appear with the same relative frequency. At its core is a measure of codon usage bias we call the Kullback–Leibler codon information bias (KL-CIB or CIB for short). Being defined in terms of conditional entropy makes KL-CIB an ideal and natural quantity for expressing a chromosome’s degree of departure from uniform synonymous codon usage. Applying the approach to a large collection of annotated bacterial chromosomes reveals three distinct groups of bacteria.