N-gram analysis of prokaryotic genomes: characteristics and predictions of genomic islands

Abstract

Bacterial genomes were shown to contain parts of various lenght that differ in base composition (GC frequency, codon usage, signature profile) and gene content (novel genes that may contribute to bacterial adaptation), that are probably aquired by horizintal gene transfer and are designated as genome islands (GI). GIs taht contain a variety of virulence factors and contribute to bacterial pahtogenecity are designated as Pathogenic islands (PAIs). In order to predict and characterize PAIs more precisely, we introduced n-gram analisis in addition to other compositional features analysis. Results obtained include: binary sequence classification into PAI candidate and others, model based PAIs prediction. Our predictions were also compared with the results of others (as in PAI DB), and impruves relative precision significantly (up to 60% as compared to PAI DB). It was demonstarted that union of all compositional features results in maximum recall up to 37%. Thus, n-gram analysis significantly impruves PAI perdictons and may be used alongside exsisting and newly developed methods.