MRPGA: Motif Detecting by Modified Random Projection Strategy and Genetic Algorithm

Detecting common patterns or motifs in a set of DNA sequences is a major task in computational biology. Recently, this task was formally formulated as a planted (I, d)-motif problem, and several instances of the problem have been posed as challenges for motif detecting algorithms. In this work, an approach of genetic algorithm using Bayesian inference is proposed to identify (I, d)-motifs, where a modified random projection strategy is applied to generate a good initial population of the genetic algorithm. Based on our method, a program called MRPGA is developed, and experimental results on simulated data show that MRPGA performs better than Random Projection and GARPS in finding weak signal motifs. We test MRPGA on realistic biological data by identifying ERE binding sites of estradiol, CRP in Escherichia coli, as well as transcription factors in E2F family. In real-data applications, MRPGA achieves superior performances comparing with MEME, MDGA, BioProsceptor and BioOptimizor.