high performance computing on graphics processing units: hgpu.org

The identification of genome-wide transcription factor binding sites is a fundamental and crucial problem to fully understand the transcriptional regulatory processes. However, the high computational cost of many motif discovery algorithms heavily constraints their application for large-scale datasets. The rapid growth of genomic sequences and gene transcription data further deteriorates the situation and establishes a strong requirement for time-efficient scalable motif discovery algorithms. The emergence of many-core architectures, typically CUDA-enabled GPUs, provides an opportunity to reduce the execution time by an order of magnitude without the loss of accuracy. In this paper, we present mCUDA-MEME, an ultrafast scalable many-core motif discovery algorithm for multiple GPUs based on the MEME algorithm. Our algorithm is implemented using a hybrid combination of the CUDA, OpenMP and MPI parallel programming models in order to harness the powerful compute capability of modern GPU clusters. At present, our algorithm supports OOPS and ZOOPS models, which are sufficient for most motif discovery applications. mCUDAMEME achieves significant speedups for the starting point search stage (and the overall execution) when benchmarked, using real datasets, against parallel MEME running on 32 CPU cores. Speedups of up to 1.4 (1.1) on a single GPU of a Fermi-based Tesla S2050 quad-GPU computing system and up to 10.8 (8.3) on the eight GPUs of a two Tesla S2050 system were observed. Furthermore, our algorithm shows good scalability with respect to dataset size and the number of GPUs (availability:https://sites.google.com/site/yongchaosoftware/mc uda-meme).