A Deterministic Finite Automaton for Faster Protein Hit Detection in BLAST

Abstract

BLAST is the most popular bioinformatics tool and is used to run millions of queries each day. However, evaluating such queries is slow, taking typically minutes on modern workstations. Therefore, continuing evolution of BLAST-by improving its algorithms and optimizations-is essential to improve search times in the face of exponentially increasing collection sizes. We present an optimization to the first stage of the BLAST algorithm specifically designed for protein search. It produces the same results as NCBI-BLAST but in around 59% of the time on Intel-based platforms; we also present results for other popular architectures. Overall, this is a saving of around 15% of the total typical BLAST search time. Our approach uses a deterministic finite automaton (DFA), inspired by the original scheme used in the 1990 BLAST algorithm. The techniques are optimized for modern hardware, making careful use of cache-conscious approaches to improve speed. Our optimized DFA approach has been integrated into a new version of BLAST that is freely available for download at http://www.fsa-blast.org/.

High Performance Protein Sequence Database Scanning on the Cell Broadband Engine

“…A major limitation when designing SPE kernels is that their local memory is only 256 kB for both instructions and data. Using default parameter for w and T the size of the lookup table used for Stage 1 by NCBI BLASTP is already around 400 kB for a query sequence of average length [5]. Therefore, we need to use an alternative data structure which requires significantly less memory.…”

“…Therefore, we are using a more memory-efficient data structure for Stage 1. The utilized data structure is a compressed deterministic finite-state automaton (DFA), which is similar to the approach used by FSA-BLAST [4,5]. The compressed DFA for w = 3 is illustrated in Fig.…”

“…nlm.nih.gov/BLAST/developer.shtml). FSA-BLAST uses an optimized sequential algorithm and is around 15% faster than NCBI-BLASTP with no loss in accuracy [4,5]. FSA-BLASTP and NCBI-BLASTP are tested on a HP workstation xw4200 with Dual-core Pentium ® 4 (P4) CPU 3 GHz, 2 GB of RAM.…”

Accelerating BLASTP on the Cell Broadband Engine

“…Therefore, we are using a more memory-efficient data structure for Stage 1. The utilized data structure is a compressed deterministic finite-state automaton (DFA), which is similar to the approach used by FSA-BLAST [3,4]. The compressed DFA for w=3 is illustrated in Figure 4.…”

Compression of finite-state automata through failure transitions

“…Deterministic finite-state automata (DFA) have applications in natural language processing (Roche and Shabes, 1997), medical data analysis (Lewis et al, 2010), network intrusion detection (Tuck et al, 2004), computational biology (Cameron et al, 2005), and other fields. Although DFA are less compact than their non-deterministic counterpart, they are easier to work with algorithmically, and their uniform membership problem, when also the language model is part of the input, can be decided in time O(|w| log |Q|), where w is the input string and Q the state space.…”