Efficient identification of transcription factor binding sites with a graph theoretic approach

Identifying transcription factor binding sites with experimental methods is often expensive and time consuming. Although many computational approaches and tools have been developed for this problem, the prediction accuracy is not satisfactory. In this paper, we develop a new computational approach that can model the relationships among all short sequence segments in the promoter regions with a graph theoretic model. Based on this model, finding the locations of transcription factor binding site is reduced to computing maximum weighted cliques in a graph with weighted edges. We have implemented this approach and used it to predict the binding sites in two organisms, Caenorhabditis elegans and mus musculus. We compared the prediction accuracy with that of the Gibbs Motif Sampler. We found that the accuracy of our approach is higher than or comparable with that of the Gibbs Motif Sampler for most of tested data and can accurately identify binding sites in cases where the Gibbs Motif Sampler has difficulty to predict their locations.