Operon Prediction by DNA Microarray: An Approach with a Bayesian Network Model

An operon is a set of genes in prokaryotes, which are transcribed to a single mRNA transcript. Although the operon organization has not yet been completely revealed even in model organisms, such as E. Coli and B. subtilis, the understanding of the operon organization is important for various transcriptome analyses and for the prediction of genes’ functions. Many methods that predict transcription units use correlation coefficients, denoted by r, between genes, which are calculated from DNA microarray data. Because the r’s distribution of operon pairs (OPs) is different from that of non-operon pairs (NOPs), OPs and NOPs can be discriminated based on r[2]. For example, Bockhorst et al. [1] proposed a method with a Bayesian network to link many kinds of observations such as spacer sizes, expression profiles, and codon usage. Tjaden et al. [3] used a hidden Markov model (HMM) to predict transcription boundaries. Although these methods utilize correlations between only adjacent genes on a single DNA strand, a pair of genes that are not immediately next to each other can be an operon pair when the members of the operon are more than two. In this report, we propose a new method for the operon prediction, which utilizes correlations between not only adjacent but also distant genes on a DNA strand.