Accelerating iterative protein sequence alignment on a heterogeneous GPU-CPU platform

Abstract

The basic local alignment search tool (BLAST) is a fundamental bioinformatics tool used widely for biological sequence similarity searching. BLAST performs compute intensive operations on growing databases of around 46 GB today, urging researchers to find accelerated BLAST solutions. GPU accelerated BLAST solutions were introduced to exploit the low cost, high performance nature of GPU computing platforms compared to CPU multi-core platforms. Most of the introduced solutions target accelerating protein BLAST (BLASTP), focusing on the most time consuming algorithm stages (i.e. seeds collection and ungapped extension). In this paper, we present a survey on various GPU accelerated BLAST solutions, comparing them with respect to the algorithm stages implemented on GPU, the interaction between GPU and CPU execution paths, the parallelization granularity applied and other effective aspects. We propose a GPU-CPU multithreaded solution accelerating PSI-BLAST. Our work addresses accelerating the iterative protein alignment tool, PSI-BLAST for its great sensitivity to distantly related protein sequences compared to BLASTP. Our solution achieves a speedup of around 4X on an NVIDIA Tesla C2070 GPU compared to the sequential NCBI PSI-BLAST.