FPGA acceleration of sequence analysis tools in bioinformatics

Abstract

With advances in biotechnology and computing power, biological data are being produced at an exceptional rate. The purpose of this study is to analyze the application of FPGAs to accelerate high impact production biosequence analysis tools. Compared with other alternatives, FPGAs offer huge compute power, lower power consumption, and reasonable flexibility. BLAST has become the de facto standard in bioinformatic approximate string matching and so its acceleration is of fundamental importance. It is a complex highly-optimized system, consisting of tens of thousands of lines of code and a large number of heuristics. Our idea is to emulate the main phases of its algorithm on FPGA. Utilizing our FPGA engine, we quickly reduce the size of the database to a small fraction, and then use the original code to process the query. Using a standard FPGA-based system, we achieved 12x speedup over a highly optimized multithread reference code. Multiple Sequence Alignment (MSA)--the extension of pairwise Sequence Alignment to multiple Sequences--is critical to solve many biological problems. Previous attempts to accelerate Clustal-W, the most commonly used MSA code, have directly mapped a portion of the code to the FPGA. We use a new approach: we apply prefiltering of the kind commonly used in BLAST to perform the initial all-pairs alignments. This results in a speedup of from 8Ox to 190x over the CPU code (8 cores). The quality is comparable to the original according to a commonly used benchmark suite evaluated with respect to multiple distance metrics. The challenge in FPGA-based acceleration is finding a suitable application mapping. Unfortunately many software heuristics do not fall into this category and so other methods must be applied. One is restructuring: an entirely new algorithm is applied. Another is to analyze application utilization and develop accuracy/performance tradeoffs. Using our prefiltering approach and novel FPGA programming models we have achieved significant speedup over reference programs. We have applied approximation, seeding, and filtering to this end. The bulk of this study is to introduce the pros and cons of these acceleration models for biosequence analysis tools.