With the rapid development of Next-Generation Sequencing(NGS)technology,genome sequencing services for clinical fields are now bringing new challenges to existing solutions.The increasing demand for alignment data pro...With the rapid development of Next-Generation Sequencing(NGS)technology,genome sequencing services for clinical fields are now bringing new challenges to existing solutions.The increasing demand for alignment data processing motivates the development of more efficient algorithms for computational genomics.The Pair-Hidden Markov Model(Pair-HMM)is one of the most popular models used to process sequence alignment.Its related Forward Algorithm(FA)is usually the key performance bottleneck of the entire variant calling workflow.While multiple previous works have been conducted in efforts to accelerate the algorithm with various levels of parallelization,it still lacks of fully utilizing the resources of heterogeneous devices,such as high-bandwidth memory and massive SIMD cores in advanced GPU.In this paper,we design a GPU-based Pari-HMM sequence alignment algorithm and conduct its implementation with holistic co-design optimizations,including efficient computational parallelization,parameter initialization,memory accessing layout,and etc.When using Nvidia Telsa V100 GPU,Our work has shown speedups of 1151x compared to the Java baseline on Intel single-core CPU and 1.47x to the previous state-of-art GPU work.展开更多
基金supported by National Key R&D Program of China under Grant 2022ZD0115304the National Natural Science Foundation of China under Grand No.62402534+1 种基金the GuangDong Basic and Applied Basic Research Foundation:2023A1515110117the Fundamental Research Funds for the Central Universities,Sun Yat-sen University:23xkjc016.
文摘With the rapid development of Next-Generation Sequencing(NGS)technology,genome sequencing services for clinical fields are now bringing new challenges to existing solutions.The increasing demand for alignment data processing motivates the development of more efficient algorithms for computational genomics.The Pair-Hidden Markov Model(Pair-HMM)is one of the most popular models used to process sequence alignment.Its related Forward Algorithm(FA)is usually the key performance bottleneck of the entire variant calling workflow.While multiple previous works have been conducted in efforts to accelerate the algorithm with various levels of parallelization,it still lacks of fully utilizing the resources of heterogeneous devices,such as high-bandwidth memory and massive SIMD cores in advanced GPU.In this paper,we design a GPU-based Pari-HMM sequence alignment algorithm and conduct its implementation with holistic co-design optimizations,including efficient computational parallelization,parameter initialization,memory accessing layout,and etc.When using Nvidia Telsa V100 GPU,Our work has shown speedups of 1151x compared to the Java baseline on Intel single-core CPU and 1.47x to the previous state-of-art GPU work.
