Exploration of alternative GPU implementations of the pair-HMMs forward algorithm

Abstract

In order to handle the massive raw data generated by next generation sequencing (NGS) platforms, GPUs are widely used by many genetic analysis tools to speed up the used algorithms. In this paper, we use GPUs to accelerate the pair-HMMs forward algorithm, which is used to calculate the overall alignment probability in many genomics analysis tools. We firstly evaluate two different implementation methods to accelerate the pair-HMMs forward algorithm according to their effectiveness on GPU platforms. Based on these two methods, we present several implementations of the pair-HMMs forward algorithm. We execute these implementations on the NVIDIA Tesla K40 card using different datasets to compare the performance. Experimental results show that the intra-task implementation has the highest throughput in most cases, achieving pure computational throughput as high as 23.56 GCUPS for synthetic datasets. On a real dataset, the inter-task implementation achieves 4.82x speedup compared with a vectorized implementation executed on a 20-core POWER8 system.

Improved GPU Implementations of the Pair-HMM Forward Algorithm for DNA Sequence Alignment

“…Previous works have tried to improve the Pair-HMM FA from different perspectives. S. Ren et al [3], [4] utilize GPU to parallel the workload and accelerate the Pair-HMM FA by increasing throughput; S. S. Banerjee [5], S. Huang [6] and L. van Dam [7] implement Pair-HMM FA on Field-Programmable Gate Array (FPGA), with optimized data forwarding strategies and special computation routines.…”

“…Next-Generation Sequencing (NGS) technology has the features of lowcost and high-performance. To further improve the efficiency and continue providing affordable analysis resources in response of the increasing availability of genome data through public databases, researchers have started investigating the possibility of deploying new algorithms on GPU [3], [4] and FPGA [5], [6], [7], [8], [9] given the fact that bioinformatics workloads could be executed in parallel given their batch processing nature.…”

Performance extraction and suitability analysis of multi- and many-core architectures for next generation sequencing secondary analysis

“…Existing efforts to accelerate PairHMM include optimization for CPU, GPGPU, and FPGA [36][37][38][39][40][41]. Similar to SW, acceleration of PairHMM can exploit both intra-task parallelism and inter-task parallelism.…”

“…A few GPU implementations are available. Among the most recent ones, Ren et al [38] report throughput of 23.56 GCUPS on Nvidia K40 GPU. Wang et al [39] report peak throughput of 34.8 GCUPS on Nvidia Titan X GPU.…”

Pair-HMM accelerator based on non-cooperative structure

“…The number of gene sequence alignment tasks reaches more than one million. Therefore, many researchers use parallel platforms [13] such as multi-core CPUs [9,14], GPUs [15,16,17,18,19] and Field Programmable Gate Arrays (FPGAs) [20,21,22,23,24,25] to develop parallelism in gene sequence alignment applications. Compared with CPUs and GPUs, FPGA chips serve as custom hardware for the computing-intensive applications with numerous computing and storage resources [26,27].…”