A Stochastic Model for DNA Sequences Using Prescribed Nucleotide and Length Distributions

A stochastic model that generates artificial DNA sequences with correlation characteristics similar to those observed in real DNA sequences is proposed. A Bernoulli-like process is used to generate patches of DNA with nucleotide content representative of coding and noncoding region. Alternating coding and noncoding DNA patches are concatenated to form the sequence where the patch length is based on sample statistics. Examples demonstrate that the nonuniform use of codons in coding regions is responsible for the often-observed period-three property. The amplitude of the correlation corresponding to the period-three property is proportional to the coding-region length and inversely proportional to the noncoding-region length. The correlation characteristics of the complete M.tuberculosis, B.subtilis, and S.cerevisiae (chromosome XI) genomes exhibit two distinct branches corresponding to period three and nonperiod-three correlations like those observed for the artificial DNA sequences