Hardware BLAST Algorithms with Multi-seeds Detection and Parallel Extension

Abstract

As one of the most widely used bio-sequence searching tools, BLAST adopts index-based approach to detect the matches between two substrings by looking up a large table and processing one match per query. In this paper, we propose a systolic array approach to detect string matches without using looking up tables. The pipelining systolic array is implemented as a multi-seeds detection and parallel extension pipeline engine to accelerate the first two stages of NCBI BLAST algorithm. Different from the index-based approach, our implementation consumes little memory resources and eliminates redundant string extensions by merging multiple adjoin seeds into a valid seed. Our FPGA implementation achieves superior performance results in both of processing element number and clock frequency over related works in the area of FPGA BLAST accelerators. The experimental results also show the speedup can reach about 17 and 48 compared to the NCBI BLASTp and TBLASTn programs for 3072-residue queries on Intel P4 CPU, respectively. Furthermore, the idea of multi-seeds detection also can be adopted in other seed-based heuristic searching applications.