深度学习在人类基因组研究中的应用与挑战^*

Applications and Challenges of Deep Learning in Human Genome Research

摘要

近年来，随着高通量组学技术的发展，人类基因组数据迎来了爆炸式增长。通过海量数据挖掘人类基因组潜在作用已成为功能基因组研究的一项挑战。尽管传统统计方法在过去小规模数据分析中取得了成功，但其在分析效率、多维度数据整合方面存在明显局限，难以满足日益增长的数据分析需求。深度学习技术的引入为该领域研究提供了一个新的思路。本文系统综述了深度学习在人类基因组研究中的应用进展。研究表明，在标记数据完备的情况下，基于判别式深度学习的计算方法，如卷积神经网络(CNN)和长短记忆神经网络(LSTM)架构在基因组变异发现任务中表现出较高的准确性和效率。而基于生成式深度学习的计算方法，如大语言深度学习模型架构(LLM)有效利用自监督预训练策略实现复杂基因组信息的整合，在功能性基因组序列注释中表现优秀。此外，本文还探讨了LLM深度学习模型在多组学数据整合及预测中的应用。未来，随着长片段、高维度数据的积累，深度学习技术有望实现更复杂混合基因组信息的整合，在人类基因组研究中发挥更加重要的作用。

Abstract

In recent years, the advent of high-throughput omics technologies has fueled an explosive growth in human genomic data. Uncovering the latent functions within this vast data has become a significant challenge in functional genomics research. While traditional statistical methods have proved successful for analyzing smaller-scale datasets in the past, they exhibit clear limitations in analytical efficiency and integrating multi-dimensional data, struggling to meet the escalating demands of contemporary genomic analysis. The introduction of deep learning (DL) technologies offers a novel paradigm for this field. This review systematically examines the advances in applying deep learning to human genomics research. Studies demonstrate that when ample labeled data is available, discriminative DL computational methods-such as Convolutional Neural Networks (CNNs) and Long Short-Term Memory networks (LSTMs)-achieve high accuracy and efficiency in genomic variant discovery tasks. Furthermore, generative DL methods, particularly Large Language Models (LLMs) leveraging self-supervised pre-training strategies, effectively integrate complex genomic information and exhibit superior performance in functional genomic sequence annotation and gene regulation studies. This review also explores the application of LLMs in multi-omics data integration and prediction. Looking ahead, the continued accumulation of long-read sequencing and high-dimensional data is expected to enable DL technologies to integrate increasingly complex and heterogeneous genomic information, playing an increasingly crucial role in human genomics research.

关键词

深度学习 / 人类基因组 / 大语言模型 / 多组学数据整合 / 综述

Key words

Deep learning / Human genomics / Large language models / Multi-omics data integration / Review

参考文献

[1] McKenna A, Hanna M, Banks E, et al. The Genome Analysis Toolkit: a MapReduce framework for analyzing next-generation DNA sequencing data[J]. Genome Res, 2010, 20(9): 1297. DOI: 10.1101/gr.107524.110.

[2] Boeva V, Popova T, Bleakley K, et al. Control-FREEC: a tool for assessing copy number and allelic content using next-generation sequencing data[J]. Bioinformatics, 2012, 28(3): 423. DOI: 10.1093/bioinformatics/btr670.

[3] Ye K, Schulz M H, Long Q, et al. Pindel: a pattern growth approach to detect break points of large deletions and medium sized insertions from paired-end short reads[J]. Bioinformatics, 2009, 25(21): 2865. DOI: 10.1093/bioinformatics/btp394.

[4] Uffelmann E, Huang Q Q, Munung N S, et al. Genome-wide association studies[J/OL]. Nat Rev Methods Primers, 2021, 1: 59. https://www.nature.com/articles/s43586-021-00056-9. DOI: 10.1038/s43586-021-00056-9.

[5] Campagna M P, Xavier A, Lechner-Scott J, et al. Epigenome-wide association studies: current knowledge, strategies and recommendations[J/OL]. Clin Epigenet, 2021, 13: 214. https://link.springer.com/article/10.1186/s13148-021-01200-8.

[6] Belton J M, McCord R P, Gibcus J H, et al. Hi-C: a comprehensive technique to capture the conformation of genomes[J]. Methods, 2012, 58(3): 268. DOI: 10.1016/j.ymeth.2012.05.001.

[7] Zhang Y, Liu T, Meyer C A, et al. Model-based analysis of ChIP-Seq (MACS)[J/OL]. Genome Biol, 2008, 9(9): R137. https://link.springer.com/article/10.1186/gb-2008-9-9-r137.

[8] LeCun Y, Bengio Y, Hinton G. Deep learning[J]. Nature, 2015, 521(7553): 436. DOI: 10.1038/nature14539.

[9] Poplin R, Chang P C, Alexander D, et al. A universal SNP and small-indel variant caller using deep neural networks[J]. Nat Biotechnol, 2018, 36(10): 983. DOI: 10.1038/nbt.4235.

[10] Luo R, Sedlazeck F J, Lam T W, et al. A multi-task convolutional deep neural network for variant calling in single molecule sequencing[J/OL]. Nat Commun, 2019, 10(1): 998. https://www.nature.com/articles/s41467-019-09025-z. DOI: 10.1038/s41467-019-09025-z.

[11] Luo R, Wong C L, Wong Y S, et al. Exploring the limit of using a deep neural network on pileup data for germline variant calling[J]. Nat Mach Intell, 2020, 2(4): 220. DOI: 10.1038/s42256-020-0167-4.

[12] Zheng Z, Li S, Su J, et al. Symphonizing pileup and full-alignment for deep learning-based long-read variant calling[J]. Nat Comput Sci, 2022, 2(12): 797. DOI: 10.1038/s43588-022-00387-x.

[13] Ji Y R, Zhou Z H, Liu H, et al. DNABERT: pre-trained Bidirectional Encoder Representations from Transformers model for DNA-language in genome[J]. Bioinformatics, 2021, 37(15): 2112. DOI: 10.1093/bioinformatics/btab083.

[14] Dalla-Torre H, Gonzalez L, Mendoza-Revilla J, et al. Nucleotide Transformer: building and evaluating robust foundation models for human genomics[J]. Nat Methods, 2025, 22(2): 287. DOI: 10.1038/s41592-024-02523-z.

[15] Hao M, Gong J, Zeng X, et al. Large-scale foundation model on single-cell transcriptomics[J]. Nat Methods, 2024, 21(8): 1481. DOI: 10.1038/s41592-024-02305-7.

[16] Sedlazeck F J, Rescheneder P, Smolka M, et al. Accurate detection of complex structural variations using single-molecule sequencing[J]. Nat Methods, 2018, 15(6): 461. DOI: 10.1038/s41592-018-0001-7.

[17] Jiang T, Liu S, Cao S, et al. Structural variant detection from long-read sequencing data with cuteSV[J]. Methods Mol Biol, 2022, 2493: 137. DOI: 10.1007/978-1-0716-2293-3_9.

[18] Lin J, Wang S, Audano P A, et al. SVision: a deep learning approach to resolve complex structural variants[J]. Nat Methods, 2022, 19(10): 1230. DOI: 10.1038/s41592-022-01609-w.

[19] Wang S, Xu T, Zhang P, et al. Population-level structural variant characterization using pangenome graphs[J]. Nat Genet, 2026, 58(3): 664. DOI: 10.1038/s41588-026-02538-6.

[20] Elrick H, Sauer C M, Espejo V J, et al. SAVANA: reliable analysis of somatic structural variants and copy number aberrations using long-read sequencing[J]. Nat Methods, 2025, 22(7): 1436. DOI: 10.1038/s41592-025-02708-0.

[21] Keskus A G, Bryant A, Ahmad T, et al. Severus detects somatic structural variation and complex rearrangements in cancer genomes using long-read sequencing[J]. Nat Biotechnol, 2026, 44(2): 247. DOI: 10.1038/s41587-025-02618-8.

[22] Vaswani A, Shazeer N, Parmar N, et al. Attention is all you need[PP/OL]. V7. arXiv(2023-08-02)[2026-07-09]. https://arxiv.org/abs/1706.03762. DOI: 10.48550/arXiv.1706.03762.

[23] Devlin J, Chang M W, Lee K, et al. BERT: pre-training of deep bidirectional Transformers for language understanding[PP/OL]. V2. arXiv(2019-05-24)[2026-07-09]. https://arxiv.org/abs/1810.04805. DOI: 10.48550/arXiv.1810.04805.

[24] Zhou Z H, Ji Y R, Liu W J, et al. DNABERT-2: efficient foundation model and benchmark for multi-species genome[PP/OL]. V2. arXiv(2024-03-18)[2026-07-09]. https://arxiv.org/abs/2306.15006. DOI: 10.48550/arXiv.2306.15006.

[25] Nguyen E, Poli M, Faizi M, et al. HyenaDNA: long-range genomic sequence modeling at single nucleotide resolution[PP/OL]. V2. arXiv(2023-11-14)[2026-07-09]. https://arxiv.org/abs/2306.15794. DOI: 10.48550/arXiv.2306.15794.

[26] Nguyen E, Poli M, Durrant M G, et al. Sequence modeling and design from molecular to genome scale with Evo[J/OL]. Science, 2024, 386(6723): eado9336. https://www.science.org/doi/10.1126/science.ado9336.

[27] Brixi G, Durrant M G, Ku J, et al. Genome modelling and design across all domains of life with Evo 2[J]. Nature, 2026, 652(8112): 1349. DOI: 10.1038/s41586-026-10176-5.

[28] Radford A, Narasimhan K, Salimans T, et al. Improving language understanding by generative pre-training[R]. San Francisco: OpenAI, 2018.

[29] Gu A, Dao T. Mamba: linear-time sequence modeling with selective state spaces[PP/OL]. V2. arXiv(2024-05-31)[2026-07-09]. https://arxiv.org/abs/2312.00752. DOI: 10.48550/arXiv.2312.00752.

[30] Avsec Ž, Latysheva N, Cheng J, et al. Advancing regulatory variant effect prediction with AlphaGenome[J]. Nature, 2026, 649(8099): 1206. DOI: 10.1038/s41586-025-10014-0.

[31] Fudenberg G, Kelley D R, Pollard K S. Predicting 3D genome folding from DNA sequence with Akita[J]. Nat Methods, 2020, 17(11): 1111. DOI: 10.1038/s41592-020-0958-x.

[32] Avsec Ž, Agarwal V, Visentin D, et al. Effective gene expression prediction from sequence by integrating long-range interactions[J]. Nat Methods, 2021, 18(10): 1196. DOI: 10.1038/s41592-021-01252-x.

[33] Schiff Y, Kao C H, Gokaslan A, et al. Caduceus: bi-directional equivariant long-range DNA sequence modeling[PP/OL]. V2. arXiv(2024-06-05)[2026-07-09]. https://arxiv.org/abs/2403.03234. DOI: 10.48550/arXiv.2403.03234.

[34] Feng H, Wu L, Zhao B, et al. Benchmarking DNA foundation models for genomic and genetic tasks[J/OL]. Nat Commun, 2025, 16(1): 10780. https://pmc.ncbi.nlm.nih.gov/articles/PMC12663285/. DOI: 10.1038/s41467-025-65823-8.

[35] Cheng W, Song Z, Zhang Y, et al. DNALONGBENCH: a benchmark suite for long-range DNA prediction tasks[J/OL]. Nat Commun, 2025, 16(1): 10108. https://pmc.ncbi.nlm.nih.gov/articles/PMC12627797/. DOI: 10.1038/s41467-025-65077-4.

[36] Ledneva D, Nuridinov M, Kuznetsov D. GENEB: Why genomic models are hard to compare[PP/OL]. V3. arXiv(2026-06-11)[ 2026-07-09]. https://arxiv.org/abs/2606.04525. DOI: 10.48550/arXiv.2606.04525.

[37] Yang H, Cole J, Li Y, et al. Omnigenbench: a modular platform for reproducible genomic foundation models benchmarking[PP/OL]. V1. arXiv(2025-05-20)[2026-07-09]. https://arxiv.org/abs/2505.14402. DOI: 10.48550/arXiv.2505.14402.

[38] Nurk S, Koren S, Rhie A, et al. The complete sequence of a human genome[J]. Science, 2022, 376(6588): 44. DOI: 10.1126/science.abj6987.

[39] Gao Z, Liu Q, Zeng W, et al. EpiGePT: a pretrained transformer-based language model for context-specific human epigenomics[J/OL]. Genome Biol, 2024, 25(1): 310. https://link.springer.com/article/10.1186/s13059-024-03449-7.

[40] Angermueller C, Lee H J, Reik W, et al. DeepCpG: accurate prediction of single-cell DNA methylation states using deep learning[J/OL]. Genome Biol, 2017, 18(1): 67. https://link.springer.com/article/10.1186/s13059-017-1189-z.

[41] Hammelman J, Gifford D K. Discovering differential genome sequence activity with interpretable and efficient deep learning[J/OL]. PLoS Comput Biol, 2021, 17(8): e1009282. https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1009282.

[42] Zhou J, Troyanskaya O. Predicting effects of noncoding variants with deep learning-based sequence model[J]. Nat Methods, 2015, 12(10): 931. DOI: 10.1038/nmeth.3547.

[43] Page L, Singh R, Jenei A, et al. CpGPT: a foundation model for DNA methylation[PP/OL]. V3. BioRxiv(2025-08-28)[2026-07-09]. https://www.biorxiv.org/content/10.1101/2024.10.24.619766v4. DOI: 10.1101/2024.10.24.619766.

[44] Tejada-Lapuerta A, Schaar AC, Gutgesell R, et al. Nicheformer: a foundation model for single-cell and spatial omics[J]. Nat Methods, 2025, 22(12): 2525. DOI: 10.1038/s41592-025-02814-z.

[45] Blampey Q, Benkirane H, Bercovici N, et al. Novae: a graph-based foundation model for spatial transcriptomics data[J]. Nat Methods, 2025, 22(12): 2539. DOI: 10.1038/s41592-025-02899-6.

[46] Bunne C, Roohani Y, Rosen Y, et al. How to build the virtual cell with artificial intelligence: priorities and opportunities[J]. Cell, 2024, 187(25): 7045. DOI: 10.1016/j.cell.2024.11.015.

[47] Thornburg Z R, Maytin A, Kwon J, et al. Bringing the genetically minimal cell to life on a computer in 4D[J]. Cell, 2026, 189(9): 2582. DOI: 10.1016/j.cell.2026.02.009.

[48] Liao WW, Asri M, Ebler J, et al. A draft human pangenome reference[J]. Nature, 2023, 617(7960): 312. DOI: 10.1038/s41586-023-05896-x.

[49] Hickey G, Monlong J, Ebler J, et al. Pangenome graph construction from genome alignments with Minigraph-Cactus[J]. Nat Biotechnol, 2024, 42(4): 663. DOI: 10.1038/s41587-023-01793-w.

[50] Schloissnig S, Pani S, Ebler J, et al. Structural variation in 1,019 diverse humans based on long-read sequencing[J]. Nature, 2025, 644(8076): 442. DOI: 10.1038/s41586-025-09290-7.

[51] Fallahpour A, Magnuson A, Gupta P, et al. BioReason: incentivizing multimodal biological reasoning within a DNA-LLM model[PP/OL]. V2. arXiv(2025-10-24)[2026-07-09]. https://arxiv.org/abs/2505.23579. DOI: 10.48550/arXiv.2505.23579.

[52] Zhang X, Yang M, Yin X, et al. DeepGene: an efficient foundation model for genomics based on pan-genome graph transformer[J]. IEEE Trans Comput Biol Bioinform, 2025, 22(6): 3143. DOI: 10.1109/TCBBIO.2025.3614354.

[53] Shipony Z, Marinov G K, Swaffer M P, et al. Long-range single-molecule mapping of chromatin accessibility in eukaryotes[J]. Nat Methods, 2020, 17(3): 319. DOI: 10.1038/s41592-019-0730-2.

基金资助

*国家自然科学基金项目(82173383)

国家自然科学基金项目(82241238)

国家自然科学基金项目(32300522)

四川大学华西医院学科卓越发展1·3·5工程项目(ZYYC23024)