Protein Sequence Motif Discovery on Distributed Supercomputer

Abstract

The motif discovery problem has gained lot of significance in biological science over the past decade. Recently, various approaches have been used successfully to discover motifs. Some of them are based on probabilistic approach and others on combinatorial approach. We follow a graph-based approach to solve this problem, in particular, using the idea of de Bruijn graphs. The de Bruijn graph has been successfully adopted in the past to solve problems such as local multiple alignment and DNA fragment assembly. The proposed algorithm harnesses the power of the de Bruijn graph to discover the conserved regions such as motifs in a protein sequence. The sequential algorithm has 70% matches of the motifs with the MEME and 65% pattern matches with the Gibbs motif sampler. The motif discovery problem is data intensive requiring substantial computational resources and cannot be solved on a single system. In this paper, we use the distributed supercomputers available on the Western Canada Research Grid (WestGrid) to implement the distributed graph based approach to the motif discovery problem and study its performance analysis. We use the available resources efficiently to distribute data among the multicore nodes in the machine and redesign the algorithm to suit the architecture. We show that a pure distributed implementation is not efficient for this problem. We develop a hybrid algorithm that uses fine grain parallelism within the nodes and coarse grain parallelism across the nodes. Experiments show that this hybrid algorithm runs 3 times faster than the pure distributed memory implementation.