A new hardware architecture for genomic and proteomic sequence alignment

Abstract

We describe a novel hardware architecture for genomic and proteomic sequence alignment which achieves a speed-up of two to three orders of magnitude over Smith-Waterman dynamic programming (DP) in hardware. In our previous papers, we introduce several features of our search algorithm, DASH, which outperforms NCII-Blast (BLAST) by an order of magnitude in software, and has better sensitivity. Indeed, DASH has been shown to have excellent sensitivity compared to Smith-Waterman. It is designed around the principle of considering genomic and proteomic sequence alignments to typically consist of regions of high homology (the diagonals) interspersed with regions of low homology. In DASH, the optimal solution consists of such diagonals joined by regions of exact DP. This is affordable due to the small area of these interconnecting regions. Accordingly, we have designed a chip which finds the diagonals and performs the inter-region DP directly in hardware. On a Xilinx Vitex II, XC2V6000, FPGA, it performs over 10/sup 12/ base comparisons/second.