Accelerating HMMER search using FPGA

Abstract

This paper describes an implementation of HMMER with FPGA. HMMER is one of the most used software tools for sensitive profile HMM (Hidden Markov Model) searches of biological sequence databases. HMMER is a very cpuintensive program. In HMMER, the Viterbi algorithm, which is a quadratic dynamic programming algorithm, is used to align a profile HMM and a protein sequence. In the profile HMM, a feedback path from the end of the model to the beginning is allowed, and this loop makes it difficult to process the Viterbi algorithm in parallel. In our approach, the alignment is calculated speculatively in parallel, and when the feedback path is selected in the alignment, the alignment is recalculated from the beginning using the fedback score. According to our experiments, the ratio that the feedback path is selected is very low, and the performance loss by the recalculation is less than a few percent. Another problem for accelerating HMMER using FPGA is the large size of the score tables required for profile HMMs. By crossing the search direction in the quadratic search space and the moving direction of each processing unit in the search space, we can minimize the size of the memory banks for storing the score tables. This optimization technique makes it possible to process all profile HMMs in a database efficiently.

Parallel accelerators for GlimmerHMM bioinformatics algorithm

“…Oliver et al in [9,10] mapped the Plan7 Viterbi algorithm, which is the basic stage of the HMMER algorithm, on a Spartan 3 FPGA achieving acceleration of up to 30x against the execution of this same stage on a high-end AMD CPU. Takagi et al in [11] propose a fully parallel implementation of the Viterbi algorithm which, when it is executed on a high end FPGA, it achieves an up to 360x speedup for certain sequences vs. the execution of the official software on a Core2Duo CPU. Chrysos et al in [12] describe the first FPGA-based implementation of the Glimmer algorithm, which is a gene finding algorithm tailored to prokaryote organisms and based on the Interpolated Markov model; this latter model is also used in the GlimmerHMM algorithm.…”

Combining execution pipelines to improve parallel implementation of HMMER on FPGA

“…More recently, an approach for hardware acceleration based on speculative execution was proposed by Takagi et al [18] and Sun et al [19]. Their idea is to take advantage of some property of the max operation, so as speculatively ignore the dependency over variable B½i, (1), since it very seldom contributes to the actual computation of M½i þ 1; k; D½i þ 1; k and I½i þ 1; k. This then results in a feedback-loop free algorithm, like [12,13], which is very easy to parallelize.…”

Accelerating ncRNA homology search with FPGAs

“…The feedback path in Plan 7 allows for the model to match multiple copies of itself in succession [14]. This feedback requires special consideration for an FPGA implementation, and so the lack of it in CP9 gives us greater design freedom.…”

“…However, given their size and structural restrictions, the primary structure alone provides sufficient sensitivity for finding protein family members. Of the efforts to accelerate HMMER on FPGAs, [14] is the most relevant. It presents a very fast FPGA implementation of Plan 7, with the ability to fit over 50 PEs on a Virtex-4.…”

“…Both of these algorithms have a number of FPGA implementations [12][13] [14] [15]. However, to our knowledge there have been no previous efforts to accelerate Infernal's entire ncRNA homology search pipeline.…”