变异识别算法与优化方法综述

doi:10.11871/jfdc.issn.2096-742X.2026.05.011

数据与计算发展前沿 ›› 2026, Vol. 8 ›› Issue (5): 134-152.

CSTR: 32002.14.jfdc.CN10-1649/TP.2026.05.011

doi: 10.11871/jfdc.issn.2096-742X.2026.05.011

变异识别算法与优化方法综述

阮佳辉^1,²(),冯仰德^1,^2,^*(),李瑞琳¹

¹ 中国科学院计算机网络信息中心，北京 100083
² 中国科学院大学，北京 100190

收稿日期:2025-10-28 出版日期:2026-10-20 发布日期:2026-10-20
通讯作者: ^*冯仰德（E-mail：ydfeng@sccas.cn）
作者简介:阮佳辉，中国科学院计算机网络信息中心，硕士研究生，主要研究方向是高性能计算，变异识别。
本文中主要承担工作为文献调研，整理分析，文章撰写。
RUAN Jiahui is a graduate student at the Computer Network Information Center, Chinese Academy of Sciences. His main research interests are high-performance computing and variant identification.
In this paper, he is responsible for literature research, organization and analysis, and article writing. E-mail： jhruan@cnic.cn|冯仰德，中国科学院计算机网络信息中心，副研究员，硕士生导师，主要研究方向为材料计算、高性能计算。
本文中主要承担工作为论文框架设计，审阅和编辑文章。
FENG Yangde is an associate professor and a master’s supervisor at the Computer Network Information Center, Chinese Academy of Sciences. His main research interests are material computation and high-performance computing.
In this paper, he is mainly responsible for the design of the paper framework, reviewing and editing the article. E-mail: ydfeng@sccas.cn
基金资助:
国家自然科学基金重大研究计划项目(92259101);中国科学院战略性先导科技专项(XDB0500101);中国科学院计算机网络信息中心基础研究基金青年项目(E4552202)

A Review of Variant Identification Algorithms and Optimization Methods

RUAN Jiahui^1,²(),FENG Yangde^1,^2,^*(),LI Ruilin¹

¹ Computer Network Information Center, Chinese Academy of Sciences, Beijing 100083, China
² University of Chinese Academy of Sciences, Beijing 100190, China

Received:2025-10-28 Online:2026-10-20 Published:2026-10-20

摘要/Abstract

摘要：

【目的】系统综述变异识别算法的应用现状与其优化方法，总结当前变异识别算法的研究进展，并展望未来的研究方向。【方法】通过深入研究近五年的变异识别工具与它们的优化方法，重点探讨变异识别工具工作流程、核心功能、准确率和性能优化方法等关键技术的研究进展。【结果】当前研究对于SNP和Indel的检测已经形成了较为完整高效的流程，性能优化形成了基于并行计算与硬件加速的方法且仍在不断完善，然而现有技术的软件通用性存在局限，对于结构变异的识别上仍存在准确率不高的问题，难以检测到所有结构变异。【结论】本文系统梳理了近五年变异识别算法工具的关键技术和优化方法。未来变异识别需聚焦结构变异识别、完善多方法协同工作的方法，解决变异识别工具通用性不足的难题。

关键词: 变异识别, 并行计算, 异构优化, FPGA优化, 深度学习

Abstract:

[Objective] This paper systematically reviews the application status and optimization methods of variant identification algorithms, summarizes the current research progress of the algorithms, and discusses the future research directions. [Methods] By conducting in-depth research on the variant identification tools and their optimization methods over the past five years, this study focuses on exploring the research progress of key technologies, such as the workflow, core functions, accuracy, and performance optimization methods of these tools. [Results] Current research has established relatively complete and efficient workflows for SNP and Indel detection, and performance optimization methods based on parallel computing and hardware acceleration have been developed and are still being improved. However, existing software has limitations in general applicability, and the accuracy of structural variant identification remains insufficient, making it difficult to detect all structural variants. [Conclusion] This paper systematically analyzes the key technologies and optimization methods of the variant identification algorithm tools over the past five years. In the future, variant identification should focus on structural variation detection, improving methods for multi-method collaborative work, and solve the problem of insufficient versatility of the variant identification tools.

Key words: variant identification, parallel computing, heterogeneous optimization, FPGA optimization, deep learning

阮佳辉, 冯仰德, 李瑞琳. 变异识别算法与优化方法综述[J]. 数据与计算发展前沿, 2026, 8(5): 134-152.

RUAN Jiahui, FENG Yangde, LI Ruilin. A Review of Variant Identification Algorithms and Optimization Methods[J]. Frontiers of Data and Computing, 2026, 8(5): 134-152, https://cstr.cn/32002.14.jfdc.CN10-1649/TP.2026.05.011.

图/表 7

图1

图2

表1

图3

表2

图4

表3

变异识别优化方法对比表"

核心优化方案	适用场景	优点	缺点	对应文献
PE环结构	中等长度序列变异分析、需灵活扩展的计算场景	灵活性和可扩展性好，便于适配不同数据规模，时间复杂度O（N*L²/P）	对于极长序列，数据传输和存储开销较大	[28]
波前计算模式	GPU加速的并行计算场景、内存受限的分析任务	减少内核内存占用，实现高效并行计算，时间复杂度O(N*L/W)	性能存在波动，对GPU内存带宽要求较高	[54]
非合作结构	高计算密度的变异分析、可扩展性要求高的场景	提高计算效率，良好的可扩展性，时间复杂度为O（N*L²/P）	对内存带宽需求较高	[55]
动态内存交换方法	GPU内存资源紧张的大规模数据处理场景	充分利用GPU内存资源，提高内存利用率，时间复杂度为O（N*L）	实现复杂度高，内存交换带来额外开销	[56]
深度学习端到端判别	复杂区域变异检测、难例变异挖掘	可自动学习变异特征，对复杂测序信号处理能力强；能有效挖掘传统算法遗漏的难例变异，时间复杂度为O(NCk²)	需大量标注数据训练模型；成本较高	[11]
扩展Manta算法	染色体结构变异检测	结构变异检测灵敏度高；可精准识别传统方法难检测的染色体大片段变化，时间复杂度为O(RL+KM²)	对短读长数据的复杂结构变异解析能力仍有局限	[21]
GATK最佳实践流程兼容与加速	现有基因组分析流程、标准化分析场景	兼容成熟流程，降低用户迁移成本；保留标准分析结果的可比性	受限于GATK原有框架，部分自定义优化难以实现；对非标准流程适配性差	[8]
大规模队列联合调用可扩展性设计	生物银行级大规模样本分析、群体遗传学研究	支持高通量分析；大规模数据处理速度快	任务调度复杂，易出现资源竞争；对硬件集群规模要求较高	[9]

表3

参考文献 56

[1]	ZHU Q L, LIU S, GAO P, et al. High-throughput sequencing technology and its application[J]. Journal of Northeast Agricultural University (English Edition), 2014, 21(3): 84-96.
[2]	李庆岗, 陶立. 高通量测序技术及其在生命科学中的应用[J]. 畜牧与饲料科学, 2012, 33(2): 25.
[3]	MASTROROSA F K, MILLER D E, EICHLER E E. Applications of long-read sequencing to Mendelian genetics[J]. Genome Medicine, 2023, 15: 42.
[4]	GNERRE S, MACCALLUM I, PRZYBYLSKI D, et al. High-quality draft assemblies of mammalian genomes from massively parallel sequence data[J]. Proceedings of the National Academy of Sciences of the United States of America, 2011, 108(4): 1513-1518.
[5]	SIRÉN J, MONLONG J, CHANG P C, et al. Pangenomics enables genotyping of known structural variants in 5202 diverse genomes[J]. Science, 2021, 374: eabg8871.
[6]	JI Y, LI Y, LIANG Y, et al. Unveiling novel genetic variants in 370 challenging medically relevant genes using the long read sequencing data of 41 samples from 19 global populations[J]. Molecular Genetics and Genomics, 2024, 299: 65.
[7]	SHAFIN K, PESOUT T, LORIG-ROACH R, et al. Nanopore sequencing and assembly of a human genome with ultra-long reads[J]. Nature Biotechnology, 2022, 40(8): 1293-1300.
[8]	MCKENNA A, HANNA M, BANKS E, et al. The Genome Analysis Toolkit: a MapReduce framework for analyzing next-generation DNA sequencing data[J]. Genome Research, 2010, 20(9): 1297-1303.
[9]	BEHERA S, CATREUX S, ROSSI M, et al. Comprehensive and accurate genome analysis at scale using DRAGEN accelerated algorithms[EB/OL]. (2024-01-02)[2025-10-10]. BioRxiv. DOI: https://doi.org/10.1101/2024.01.02.573821.
[10]	EDGE P, BANSAL V. Longshot enables accurate variant calling in diploid genomes from single-molecule long read sequencing[J]. Nature Communications, 2019, 10: 4660.
[11]	POPLIN R, CHANG P C, ALEXANDER D, et al. A universal SNP and small-indel variant caller using deep neural networks[J]. Nature Biotechnology, 2018, 36: 983-987.
[12]	PARK J, COOK D E, CHANG P C, et al. Accurate somatic small variant discovery for multiple sequencing technologies with DeepSomatic[J]. Nature Biotechnology, 2025.
[13]	GUILLE A, HADJADJ D, BLANC J F, et al. A benchmarking study of individual somatic variant callers and voting-based ensembles for whole-exome sequencing[J]. Briefings in Bioinformatics, 2024, 26(1): bbae697.
[14]	DAVIS J, GOADRICH M. The relationship between Precision-Recall and ROC curves[C]//Proceedings of the 23rd International Conference on Machine Learning. New York: ACM, 2006: 233-240.
[15]	CHENG H, CONCEPCION G T, FENG X, et al. Haplotype-resolved de novo assembly using phased assembly graphs with hifiasm[J]. Nature Methods, 2021, 18: 170-175.
[16]	连帅彬, 郭东亮, 戴宪华. 基因结构变异检测方法综述[J]. 现代生物医学进展, 2012, 12(18): 3577-3580.
[17]	GAO Y, LIU Y, MA Y, et al. GraphAligner: rapid and versatile sequence-to-graph alignment[J]. Genome Biology, 2019, 20: 207.
[18]	ABYZOV A, URBAN A E, SNYDER M, et al. CNVnator: an approach to discover, genotype, and characterize typical and atypical CNVs from family and population genome sequencing[J]. Genome Research, 2011, 21(6): 974-984.
[19]	JIANG T, LIU Y Z, JIANG Y, et al. Long-read-based human genomic structural variation detection with cuteSV[J]. Genome Biology, 2020, 21(1): 189.
[20]	LAYER R M, CHIANG C, QUINLAN A R, et al. LUMPY: a probabilistic framework for structural variant discovery[J]. Genome Biology, 2014, 15: R84.
[21]	CHEN X, SCHULZ-TRIEGLAFF O, SHAW R, et al. Manta: rapid detection of structural variants and indels for germline and cancer sequencing applications[J]. Bioinformatics, 2016, 32(8): 1220-1222.
[22]	RAUSCH T, ZICHNER T, SCHLATTL A, et al. DELLY: structural variant discovery by integrated paired-end and split-read analysis[J]. Bioinformatics, 2012, 28(18): i333-i339.
[23]	ZOOK J, CHAPMAN B, WANG J, et al. Integrating human sequence data sets provides a resource of benchmark SNP and indel genotype calls[J]. Nature Biotechnology, 2014, 32: 246-251
[24]	LIU Z, XIE Z, LI M. Comprehensive and deep evaluation of structural variation detection pipelines with third-generation sequencing data[J]. Genome Biology, 2024, 25: 188.
[25]	RAJABY R, SUNG W K. SurVIndel2: improving copy number variant calling from next-generation sequencing using hidden split reads[J]. Nature Communications, 2024, 15: 10473.
[26]	NARDONE G G, CORTESE A, BRESADOLA L, et al. A hitchhiker guide to structural variant calling: a comprehensive benchmark through different sequencing technologies[J]. Biomedicines, 2025, 13(8): 1949.
[27]	VAN DER AUWERA G A, CARNEIRO M O, HARTL C, et al. From FastQ data to high-confidence variant calls: the Genome Analysis Toolkit best practices pipeline[J]. Current Protocols in Bioinformatics, 2013, 43: 11.10. 1-11.10.33.
[28]	HUANG S, MANIKANDAN G J, RAMACHANDRAN A, et al. Hardware acceleration of the pair-HMM algorithm for DNA variant calling[C]//Proceedings of the 2017 ACM/SIGDA International Symposium on Field-Programmable Gate Arrays. ACM, 2017: 275-284.
[29]	PEVZNER P A, TANG H, WATERMAN M S. An Eulerian path approach to DNA fragment assembly[J]. Proceedings of the National Academy of Sciences of the United States of America, 2001, 98(17): 9748-9753.
[30]	DURBIN R, EDDY S, KROGH A, et al. Biological Sequence Analysis: Probabilistic Models of Proteins and Nucleic Acids[M]. Cambridge: Cambridge University Press, 1998.
[31]	KIM S, SCHEFFLER K, HALPERN A L, et al. Strelka2: fast and accurate calling of somatic and germline variants[J]. Nature Methods, 2018, 15(9): 761-763.
[32]	CIBULSKIS K, LAWRENCE M S, CARTER S L, et al. Sensitive detection of somatic point mutations in impure and heterogeneous cancer samples[J]. Nature Biotechnology, 2013, 31(3): 213-219.
[33]	BENJAMIN D, SATO T, CIBULSKIS K, et al. Calling Somatic SNVs and Indels with Mutect2[EB/OL]. (2019-12-02)[2025-10-10]. BioRxiv. DOI: https://doi.org/10.1101/861054.
[34]	NARZISI G, CORVELO A, ARORA K, et al. Genome-wide somatic variant calling using localized colored de Bruijn graphs[J]. Communications Biology, 2018, 1(1): 20.
[35]	LIN M F, RODEH O, PENN J, et al. GLnexus: joint variant calling for large cohort sequencing[EB/OL]. (2018-06-11)[2025-10-01]. BioRxiv. DOI: https://doi.org/10.1101/343970.
[36]	SMITH T F, WATERMAN M S. Identification of common molecular subsequences[J]. Journal of Molecular Biology, 1981, 147(1): 195-197.
[37]	MASOOD D, REN L, NGUYEN C, et al. Evaluation of somatic copy number variation detection by NGS technologies and bioinformatics tools on a hyper-diploid cancer genome[J]. Genome Biology, 2024, 25(1): 163.
[38]	ZOOK J, CATOE D, MCDANIEL J, et al. Extensive sequencing of seven human genomes to characterize benchmark reference materials[J]. Scientific Data, 2016, 3: 160025.
[39]	ZHENG Z, LI S, SU J, et al. Symphonizing pileup and full-alignment for deep learning-based long-read variant calling[J]. Nature Computational Science, 2022, 2(12): 797-803.
[40]	FUKUSHIMA K. Neocognitron: a self-organizing neural network model for a mechanism of pattern recognition unaffected by shift in position[J]. Biological Cybernetics, 1980, 36(4): 193-202.
[41]	SHAFIN K, PESOUT T, CHANG P C, et al. Haplotype-aware variant calling with PEPPER-Margin-De-epVariant enables high accuracy in nanopore long-reads[J]. Nature Methods, 2021, 18: 1322-1332.
[42]	ASRI M, CHANG P C, MIER J C, et al. Pangenome-aware DeepVariant[EB/OL]. (2025-6-5)[2025-9-20].BioRxiv. DOI:https://doi.org/10.1101/2025.06.05.65-7102.
[43]	HE K, ZHANG X, REN S, et al. Deep residual learning for image recognition[C]//Proc. IEEE CVPR. 2016:770-778.
[44]	SUPERNAT A, VIDARSSON O V, STEEN V M, et al. Comparison of three variant callers for human whole genome sequencing[J]. Scientific Reports, 2018, 8(1): 17851.
[45]	CHEN N C, KOLESNIKOV A, GOEL S, et al. Improving variant calling using population data and deep learning[J]. BMC Bioinformatics 2023, 24(197).
[46]	ZHU T, VATS P, ONKEN S, et al. Parabricks: GPU Accelerated Universal Pan-Instrument Genomics An-alysis Software Suite[EB/OL]. (2025-08-04) [2025-10-27]. BioRxiv. DOI: https://doi.org/10.1101/2025.07.23.666378.
[47]	FARJOUN Y, GAUTHIER L, FRIEDMAN S F, et al. A comprehensive review of deep learning-based variant calling methods[J]. Briefings in Functional Genomics, 2024.
[48]	MUTLU O, ALSER M, SINGH G, et al. Accelerating genome analysis: a primer on an ongoing journey[J]. IEEE Micro, 2020, 40(5): 65-75.
[49]	SCHMIDT B, KALLENBORN F, WICHMANN A, et al. gpuPairHMM: High-speed Pair-HMM Forward Algorithm for DNA Variant Calling on GPUs[J]. ArXiv preprint ArXiv:2411.11547v1, 2024.
[50]	O’CONNELL K A, YOSUFZAI Z B, CAMPBELL R A, et al. Accelerating genomic workflows using NVIDIA Parabricks[J]. BMC Bioinformatics 2023, 24(1): 221.
[51]	REN S, BERTELS K, AL-ARS Z. GPU-Accelerated GATK HaplotypeCaller with Load-Balanced Multi-Process Optimization[C]//2017 IEEE 17th International Conference on Bioinformatics and Bioengineering. IEEE, 2017: 497-502.
[52]	University of Toronto. Lecture 8:State-Space Models-Forward Algorithm (HMM)[EB/OL]. (2022-01-01)[2025-10-27]. DOI: https://www.cs.toronto.edu/-duvenaud/courses/csc412/lectures/08-hmm.pdf.
[53]	FREED D, ALDANA R, WEBER J A, et al. The sentieon genomics tools-a fast and accurate solution to variant calling from next-generation sequence data[EB/OL]. BioRxiv, (2017-3-10)[2025-12-08]. DOI:10.1101/115717.
[54]	REN S, BERTELS K, AL-ARS Z. Efficient acceleration of the Pair-HMMs Forward Algorithm for GA-TK HaplotypeCaller on graphics processing units[J]. Evolutionary Bioinformatics, 2018, 14: 117693431-8760543.
[55]	WANG P, LEI Y, DOU Y. Pair-HMM accelerator based on non-cooperative structure[J]. IEICE Electronics Express, 2019, 16(15): 20190402.
[56]	BRANCHINI B, ZENI A, SANTAMBROGIO M D. A methodology for Accelerating Variant Calling on GPU[C]//SC21: The International Conference for High Performance Computing, Networking, Storage and Analysis. IEEE, 2021.

变异类型	准确性指标（HG002样本）	对比工具优势
SNV	F1值99.85%	错误数比DeepVariant+BWA减少58.7%，比GATK+BWA减少83%
indel（2-50 bp）	F1值99.80%	在CMRG区域F1值98.64%，比Giraffe+DeepVariant高0.54%
结构变异（插入）	F1值76.90%	是Manta（34.90%）的2.2倍，Delly（4.70%）的16.4倍
结构变异（缺失≥50 bp）	F1值82.60%	比Manta（70.80%）高11.8%，比Lumpy（66.80%）高15.8%
CNV（1-5 kb）	F1值92.60%	是CNVnator（39.20%）的2.4倍
STR（重叠区域）	F1值96.72%	比GangSTR（69.86%）高26.86%

工具名称	测序平台	覆盖度	使用样本	SNP（精度/召回率/F1分数）	Indel（精度/召回率/F1分数）	优势区域	参考文献
Clair3	ONT	50×	HG003样本	99.67%/ 99.60%/-	90.86% /64.73%/-	复杂区域，低覆盖、大规模检测	[39]
PEPPER-Margin-DeepVariant	ONT	50×	HG003样本	99.61%/ 99.63%/-	87.62%/ 57.42%/-	首次实现ONT-SNP精度超Illumina短读段，特定复杂区域，单倍型定相	[41]
DeepVariant	Illumina	35×	HG003样本	99.82%/ 99.38%/-	99.74%/ 99.39%/	Indel检测精度和召回率最优；碱基替换错误分布更随机，无明显偏向性	[44] [45]
DeepVariant	Illumina	30×	HG001样本	99.23%/ 96.89%/ 98.05%	96.74%/ 92.11%/ 94.37%	Indel检测精度和召回率最优；碱基替换错误分布更随机，无明显偏向性	[44] [45]
GATK 4.0	Illumina	30×	HG001样本	98.61%/ 97.04%/ 97.82%	94.83%/ 86.02%/ 90.21%	SNP召回率略高于DeepVariant；群体规模分析中常用的变异质量归一化支持更成熟	[44]

变异识别算法与优化方法综述

A Review of Variant Identification Algorithms and Optimization Methods

RichHTML

PDF

可视化

摘要/Abstract

引用本文

使用本文

图/表 7

参考文献 56

相关文章 15

编辑推荐

Metrics

本文评价

[1]	伍清林, 徐瑾昊, 孙兴亮, 郭学军, 冯敏. 基于地形感知混合注意力网络的数字高程模型超分辨率重建[J]. 数据与计算发展前沿, 2026, 8(5): 122-133.
[2]	王博, 徐玉灿, 高继涵, 佘伟光, 梁存霞, 靳冰, 王鑫春, 周艳聪, 孙宝. 基于边缘检测的气管插管内窥镜图像气道边界提取[J]. 数据与计算发展前沿, 2026, 8(5): 34-47.
[3]	靳润锋, 梁文昊, 马英晋, 任鹏举, JAMES P Lewis, 金钟. 分子动力学模拟软件FIREBALL的异构加速优化[J]. 数据与计算发展前沿, 2026, 8(4): 123-135.
[4]	张梓豪, 张小瑞, 孙伟, 周石煜. 一种用于面部视频抑郁水平评估的多尺度时间差分网络[J]. 数据与计算发展前沿, 2026, 8(4): 149-161.
[5]	林予松, 梁振羽, 赵哲. PMDeUNet：一个用于高效分割腹部医学图像的轻量级网络模型[J]. 数据与计算发展前沿, 2026, 8(4): 162-178.
[6]	张思扬, 胡林. 融合标签语义的农业科学数据层次多标签分类方法研究[J]. 数据与计算发展前沿, 2026, 8(4): 246-256.
[7]	周玉明, 张一鸣, 刘圆圆, 黄山. 遥感卫星图像船舶检测样本数据集研究综述[J]. 数据与计算发展前沿, 2026, 8(2): 215-226.
[8]	王兆滨, 王睿, 吕永科, 张耀南. 基于自适应语义连接和感知注意力的沙漠分割方法[J]. 数据与计算发展前沿, 2026, 8(2): 25-39.
[9]	韩立钦, 李龙园, 王钰康, 张耀南, 常盟盟, 潘清元. 基于频域引导特征的泥石流灾后遥感影像语义分割网络构建研究[J]. 数据与计算发展前沿, 2026, 8(2): 54-65.
[10]	潘语泉,袁得嵛,贾源,王安然. 基于融合特征的VGAT-VGAN跨社交网络身份关联算法[J]. 数据与计算发展前沿, 2026, 8(1): 103-118.
[11]	蔡毅,王晓宾,陈蕊丽,韩珣. 基于笔迹的书写者性别与年龄检测研究综述[J]. 数据与计算发展前沿, 2026, 8(1): 129-147.
[12]	万萌, 何洪林, 任小丽, 聂宁明, 曹荣强, 王宗国, 李凯, 王晓光, 王彦棡, 王珏, 高超. 基于工作流的陆地生态系统碳循环实时同化预测系统[J]. 数据与计算发展前沿, 2026, 8(1): 168-182.
[13]	邓绎如,何洪波,王英,王闰强. 一种基于情感分析的网络舆论倾向性检测方法[J]. 数据与计算发展前沿, 2026, 8(1): 91-102.
[14]	杨沁蒙, 聂宁明, 周纯葆, 王彦棡. 基于深度学习的Taylor杆碰撞数值模拟算法[J]. 数据与计算发展前沿, 2025, 7(6): 101-110.
[15]	杜振鹏, 徐建良, 张先轶, 黄强. 基于GPU的FFT高性能算法库的实现和优化[J]. 数据与计算发展前沿, 2025, 7(6): 124-135.