深度学习在林木全基因组选择中的应用与挑战：从高维特征到多目标优化

Deep learning for genomic selection in forest trees: applications and challenges from high-dimensional features to multi-objective optimization

摘要

林木育种具有周期长、杂合度高以及受基因型与环境互作影响显著等特点，传统全基因组选择（GS）模型在处理高维小样本、复杂非加性效应时存在局限。深度学习凭借多层神经网络的非线性建模能力，为解析基因型与表型之间的复杂关系提供了新途径。该技术在作物 GS 中已广泛应用，并衍生出多种融合注意力机制和轻量化设计的模型。此外，多组学与环境数据的联合建模，以及面向多性状协同优化的选择指数构建，正逐渐成为研究热点。然而，针对林木育种中长周期环境互作、高杂合基因组背景及多目标选择需求，深度学习技术的适配性与整合框架尚缺乏系统梳理。本文综述了深度学习在林木 GS 中的应用，主要内容包括：（1）基因组输入特征由单核苷酸多态性向 k-mer 及图泛基因组节点类型的演变及其计算挑战；（2）多组学与环境数据的建模方法；（3）主流深度学习模型的结构特点、适用场景及超参数调优策略；（4）从单性状基因组估计育种值预测到多性状选择指数的理论演进。本文还分析了当前面临的数据稀疏、模型过拟合和可解释性不足等问题，指出迁移学习、半监督学习以及融合生物学先验的机制建模是潜在的解决方向。最后，展望了高通量表型技术与深度学习的进一步融合，提出构建集多组学数据管理、自动化分析流程和育种决策支持于一体的智慧育种平台，以推动林木育种向全基因组智能设计转变。

Abstract

Forest tree breeding is characterized by long cycles, high heterozygosity, and significant genotype-by-environment interactions. Traditional genomic selection (GS) models face limitations when dealing with high-dimensional small-sample data and complex non-additive effects. Deep learning, with its nonlinear modeling capability based on multi-layer neural networks, offers new approaches for dissecting the complex relationships between genotype and phenotype. This technology has been widely applied in crop GS, and various models incorporating attention mechanisms and lightweight designs have been developed. In addition, joint modeling of multi-omics and environmental data, as well as the construction of selection indices for multi-trait synergistic optimization, are emerging as research hotspots. However, for forest tree breeding-characterized by long-term environmental interactions, highly heterozygous genomic backgrounds, and multi-objective selection demands-the adaptability of deep learning techniques and their integrative frameworks have not been systematically reviewed. This paper reviews the applications of deep learning in forest tree GS. The main contents include: (1) the evolution of genomic input features from single nucleotide polymorphisms to k-mer and graphical pan-genome node types, along with associated computational challenges; (2) modeling methods for multi-omics and environmental data; (3) structural characteristics, applicable scenarios, and hyperparameter tuning strategies of mainstream deep learning models; (4) the theoretical progression from single-trait genomic estimated breeding value prediction to multi-trait selection indices. This review also analyzes current major challenges, including data sparsity, model overfitting, and insufficient interpretability, and identifies transfer learning, semi-supervised learning, and mechanism-guided modeling incorporating biological priors as potential solutions. Finally, we envision the further integration of high-throughput phenotyping with deep learning and propose the construction of a smart breeding platform that integrates multi-omics data management, automated analysis pipelines, and breeding decision support, thereby facilitating the transformation of forest tree breeding toward genome-wide intelligent design.

关键词

林木育种 / 全基因组选择 / 深度学习 / 多组学整合 / 选择指数 / 智慧育种

Key words

forest tree breeding / genomic selection / deep learning / multi-omics integration / selection index / intelligent breeding

参考文献

[1] Feng J J, Dan X M, Cui Y K, et al. Integrating evolutionary genomics of forest trees to inform future tree breeding amid rapid climate change[J].Plant Communications, 2024, 5(10): 101044.

[2] Borthakur D, Busov V, Cao X H, et al. Current status and trends in forest genomics[J].Forestry Research, 2022, 2: 11.

[3] 杜庆章, 战鹏宇, 李鹏, 等. 基因组选择研究进展及其在林木中的发展趋势[J].北京林业大学学报, 2020, 42(11): 1-8.

[4] Du Q Z, Zhan P Y, Li P, et al. Advances in genomic selection and its development trend in forest[J].Journal of Beijing Forestry University, 2020, 42(11): 1-8.

[5] Sharma U, Sankhyan H P, Kumari A, et al. Genomic selection: a revolutionary approach for forest tree improvement in the wake of climate change[J].Euphytica, 2023, 220(1): 9.

[6] Muranty H, Jorge V, Bastien C, et al. Potential for marker-assisted selection for forest tree breeding: lessons from 20 years of MAS in crops[J].Tree Genetics & Genomes, 2014, 10(6): 1491-1510.

[7] Meuwissen T H, Hayes B J, Goddard M E. Prediction of total genetic value using genome-wide dense marker maps[J].Genetics, 2001, 157(4): 1819-1829.

[8] 刘策, 孟焕文, 程智慧. 植物全基因组选择育种技术原理与研究进展[J].分子植物育种, 2020, 18(16): 5335-5342.

[9] Liu C, Meng H W, Cheng Z H. Plant genome-wide selection breeding technical principle and research progress[J].Molecular Plant Breeding, 2020, 18(16): 5335-5342.

[10] 张苗苗, 王军辉, 卢楠, 等. 林木全基因组选择研究现状和应用[J].世界林业研究, 2021, 34(4): 26-32.

[11] Zhang M M, Wang J H, Lu N, et al. Research progress and application of whole genome selection in forest tree breeding[J].World Forestry Research, 2021, 34(4): 26-32.

[12] 朱嵊, 黄敏仁. 基因组选择在林木遗传育种研究中的进展与展望[J].林业科学, 2020, 56(11): 176-186.

[13] Zhu S, Huang M R. Recent advances and prospect of the genomic selection in forest genetics and tree breeding[J].Scientia Silvae Sinicae, 2020, 56(11): 176-186.

[14] 贾宏霞, 刘在霞, 周乐, 等. 基因组选择在肉牛中的研究进展[J].畜牧兽医学报, 2024, 55(9): 3757-3768.

[15] Jia H X, Liu Z X, Zhou L, et al. Research progress of genomic selection in beef cattle[J].Acta Veterinaria et Zootechnica Sinica, 2024, 55(9): 3757-3768.

[16] 谈成, 边成, 杨达, 等. 基因组选择技术在农业动物育种中的应用[J].遗传, 2017, 39(11): 1033-1045.

[17] Tan C, Bian C, Yang D, et al. Application of genomic selection in farm animal breeding[J].Hereditas, 2017, 39(11): 1033-1045.

[18] 倪世恒, 王子轶, 谢鑫峰, 等. 基因组选择技术在畜禽育种中的应用研究进展[J].中国畜牧杂志, 2024, 60(6): 95-101.

[19] Ni S H, Wang Z Y, Xie X F, et al. Advances in the application of genomic selection technology in livestock breeding[J].Chinese Journal of Animal Science, 2024, 60(6): 95-101.

[20] 陈皖强, 樊艳凤, 唐修君, 等. 基因组选择模型及其在家禽育种中的应用[J].中国家禽, 2023, 45(5): 100-106.

[21] Chen W Q, Fan Y F, Tang X J, et al. Genome selection model and its application in poultry breeding[J].China Poultry, 2023, 45(5): 100-106.

[22] 李棉燕, 王立贤, 赵福平. 机器学习在动物基因组选择中的研究进展[J].中国农业科学, 2023, 56(18): 3682-3692.

[23] Li M Y, Wang L X, Zhao F P. Research progress on machine learning for genomic selection in animals[J].Scientia Agricultura Sinica, 2023, 56(18): 3682-3692.

[24] 李广, 魏天娇, 孟凡凡, 等. 大豆基因组选择与基因组选配育种的现状与展望[J].吉林农业大学学报, 2025, 47(5): 761-771.

[25] Li G, Wei T J, Meng F F, et al. Current situation and prospects of soybean genomic selection and genomic mating breeding[J].Journal of Jilin Agricultural University, 2025, 47(5): 761-771.

[26] 齐新捧, 于洋, 张奇, 等. 全基因组选择技术在大豆育种中的应用与展望[J].大豆科学, 2025, 44(6): 129-138.

[27] Qi X P, Yu Y, Zhang Q, et al. Genomic selection in soybean breeding: progress and prospects[J].Soybean Science, 2025, 44(6): 129-138.

[28] 张敖, 关媛, 张学才, 等. 玉米全基因组选择育种研究进展[J].上海农业学报, 2023, 39(5): 13-18.

[29] Zhang A, Guan Y, Zhang X C, et al. Research progress in genomic selection breeding of maize[J].Acta Agriculturae Shanghai, 2023, 39(5): 13-18.

[30] 王欣, 徐一亿, 徐扬, 等. 作物全基因组选择育种技术研究进展[J].生物技术通报, 2024, 40(3): 1-13.

[31] Wang X, Xu Y Y, Xu Y, et al. Research progress in genomic selection breeding technology for crops[J].Biotechnology Bulletin, 2024, 40(3): 1-13.

[32] 曹士亮, 张学才, 张建国, 等. 全基因组选择技术在玉米育种中的应用[J].玉米科学, 2025, 33(6): 1-9.

[33] Cao S L, Zhang X C, Zhang J G, et al. Application of genomic selection in maize breeding[J].Journal of Maize Sciences, 2025, 33(6): 1-9.

[34] 孙连军, 王宏畅, 方婷, 等. 基因组选择技术在作物育种中的应用与进展[J].中国农业大学学报, 2026, 31(4): 1-12.

[35] Sun L J, Wang H C, Fang T, et al. Applications and advances of genomic selection in crop breeding[J].Journal of China Agricultural University, 2026, 31(4): 1-12.

[36] 郭敏杰, 邓丽, 苗建利, 等. 基于全基因组选择的高产花生选育方法[J].中国油料作物学报, 2024, 46(3): 697-702.

[37] Guo M J, Deng L, Miao J L, et al. A method of peanut breeding of large pod and high yield based on genomic selection[J].Chinese Journal of Oil Crop Sciences, 2024, 46(3): 697-702.

[38] Resende M D V, Resende M F R, Sansaloni C P, et al. Genomic selection for growth and wood quality in Eucalyptus: capturing the missing heritability and accelerating breeding for complex traits in forest trees[J].New Phytologist, 2012, 194(1): 116-128.

[39] Resende M R, Muñoz P, Resende M V, et al. Accuracy of genomic selection methods in a standard data set of loblolly pine (Pinus taeda L.) [J].Genetics, 2012, 190(4): 1503-1510.

[40] Chen Z Q, Klingberg A, Hallingbäck H R, et al. Preselection of QTL markers enhances accuracy of genomic selection in Norway spruce[J].BMC Genomics, 2023, 24(1): 147.

[41] Guo C C, Yin T M, Wu H T, et al. Genomic selection with GWAS-identified QTL markers enhances prediction accuracy for quantitative traits in poplar (Populus deltoides) [J].Communications Biology, 2025, 8(1): 1242.

[42] Zhou X, Zhang L, Zhang M, et al. Genomic selection for growth and wood properties in multi-generation hybrid populations of Populus deltoides [J].Horticulture Research, 2025, 12(9): uhaf165.

[43] 郭臣臣, 李旗, 李思缘, 等. 杨树杂交子代苗期动态生长性状的全基因组选择[J].林业科学, 2026, 62(1): 32-41.

[44] Guo C C, Li Q, Li S Y, et al. Genomic selection for dynamic growth traits during the seedling stage of poplar hybrid population[J].Scientia Silvae Sinicae, 2026, 62(1): 32-41.

[45] Zhu J, Liu Q, Diao S, et al. Development of a 101.6K liquid-phased probe for GWAS and genomic selection in pine wilt disease-resistance breeding in masson pine[J].The Plant Genome, 2025, 18(1): e70005.

[46] Wong C K, Bernardo R. Genomewide selection in oil palm: increasing selection gain per unit time and cost with small populations[J].Theoretical and Applied Genetics, 2008, 116(6): 815-824.

[47] Li J, Luo Y, Zhang R, et al. Decoding hybrid origins and genetic architecture of leaf traits variation in camellia via high-density 21K SNP array for genomic prediction[J].Horticulture Research, 2025, 12(11): uhaf221.

[48] Han F, Gao M, Zhao Y, et al. Improving genomic selection accuracy using a dual-path convolutional neural network framework: a terpenoid case study[J].New Phytologist, 2026, 249(2): 961-974.

[49] Hayatgheibi H, Hallingbäck H R, Gezan S A, et al. Cross-generational genomic prediction of Norway spruce (Picea abies) wood properties: an evaluation using independent validation[J].BMC Genomics, 2025, 26(1): 680.

[50] Varona L, Legarra A, Toro M A, et al. Non-additive effects in genomic selection[J].Frontiers in Genetics, 2018, 9: 78.

[51] Wang K, Abid M A, Rasheed A, et al. DNNGP, a deep neural network-based method for genomic prediction using multi-omics data in plants[J].Molecular Plant, 2023, 16(1): 279-293.

[52] Montesinos-Lopez A, Crespo-Herrera L, Dreisigacker S, et al. Deep learning methods improve genomic prediction of wheat breeding[J].Frontiers in Plant Science, 2024, 15: 1324090.

[53] Zhang T, Sun X, Li J, et al. Application and development prospect of genomic selection breeding in coniferous trees[J].Planta, 2025, 262(6): 133.

[54] Zhang X, Chen S, Liu X, et al. A k-mer-based GWAS approach empowering gene mining in polyploids [J/OL].Preprint at Research Square, 2025 [2026−05−01].https://doi.org/10.21203/rs.3.rs-7347406/v1.

[55] Zhang Y, Wang Y, Wu T, et al. NodeGWAS: leveraging graph pangenomes for sensitive and accurate association analysis in diverse diploid and polyploid species[J].Plant Communications, 2026: 101835.

[56] Lemay M A, De Ronne M, Bélanger R, et al. k-mer-based GWAS enhances the discovery of causal variants and candidate genes in soybean[J].The Plant Genome, 2023, 16(4): e20374.

[57] Roberts M, Josephs E B. Previously unmeasured genetic diversity explains part of lewontin’s paradox in a k-mer-based meta-analysis of 112 plant species [J/OL].Preprint at Research Square, 2024 [2026−05−01]. doi:10.1101/2024.05.17.594778.

[58] Zhang Z, Liu D, Li B, et al. A k-mer-based pangenome approach for cataloging seed-storage-protein genes in wheat to facilitate genotype-to-phenotype prediction and improvement of end-use quality[J].Molecular Plant, 2024, 17(7): 1038-1053.

[59] He C, Washburn J D, Schleif N, et al. Trait association and prediction through integrative k-mer analysis[J].The Plant Journal, 2024, 120(2): 833-850.

[60] Shi T, Zhang X, Hou Y, et al. The super-pangenome of populus unveils genomic facets for its adaptation and diversification in widespread forest trees[J].Molecular Plant, 2024, 17(5): 725-746.

[61] Fang Y, Xiao X, Lin J, et al. Pan-genome and phylogenomic analyses highlight hevea species delineation and rubber trait evolution[J].Nature Communications, 2024, 15(1): 7232.

[62] Montesinos-López O A, Montesinos-López A, Mosqueda-González B A, et al. Genomic prediction powered by multi-omics data[J].Frontiers in Genetics, 2025, 16: 636438.

[63] Robinson H, Robles-Zazueta C A, Voss-Fels K P. Accelerating perennial crop improvement via multi-omics-based predictive breeding[J].The Plant Genome, 2025, 18(4): e70058.

[64] Amin A, Zaman W, Park S. Harnessing multi-omics and predictive modeling for climate-resilient crop breeding: from genomes to fields[J].Genes, 2025, 16(7): 809.

[65] Tsega A, Mullualem D. Machine learning for multi-omics data integration in crop improvement: a systematic review[J].BMC Bioinformatics, 2026, 27(1): 81.

[66] Liu X, Wang M, Qin J, et al. GbyE: an integrated tool for genome widely association study and genome selection based on genetic by environmental interaction[J].BMC Genomics, 2024, 25(1): 386.

[67] Brault C, Conley E J, Read A C, et al. Improving genomic prediction for plant disease using environmental covariates[J].Plant Methods, 2025, 21(1): 114.

[68] Thingujam D, Gouli S, Cooray S P, et al. Climate-resilient crops: integrating AI, multi-omics, and advanced phenotyping to address global agricultural and societal challenges[J].Plants, 2025, 14(17): 2699.

[69] Montesinos-López O A, Montesinos-López A, Pérez-Rodríguez P, et al. A review of deep learning applications for genomic selection[J].BMC Genomics, 2021, 22(1): 19.

[70] Cecil R M, Sugden L A. On convolutional neural networks for selection inference: revealing the effect of preprocessing on model learning and the capacity to discover novel patterns[J].PLoS Computational Biology, 2023, 19(11): e1010979.

[71] Wang H, Yan S, Wang W, et al. Cropformer: an interpretable deep learning framework for crop genomic prediction[J].Plant Communications, 2025, 6(3): 101223.

[72] Deng P, Liu K, Zhou M, et al. DPCformer: an interpretable deep learning model for genomic prediction in crops [J/OL].Preprint at Research Square, 2025 [2026−05−01].https://doi.org/10.48550/arXiv.2510.08662.

[73] Ma W, Qiu Z, Song J, et al. A deep convolutional neural network approach for predicting phenotypes from genotypes[J].Planta, 2018, 248(5): 1307-1318.

[74] Gao P, Zhao H, Luo Z, et al. SoyDNGP: a web-accessible deep learning framework for genomic prediction in soybean breeding[J].Briefings in Bioinformatics, 2023, 24(6): bbad349.

[75] Wu C, Zhang Y, Ying Z, et al. A transformer-based genomic prediction method fused with knowledge-guided module[J].Briefings in Bioinformatics, 2024, 25(1): bbad438.

[76] Ma X, Wang H, Wu S, et al. DeepCCR: large-scale genomics-based deep learning method for improving rice breeding[J].Plant Biotechnology Journal, 2024, 22(10): 2691.

[77] Yao Z, Yao M, Wang C, et al. GEFormer: a genotype-environment interaction-based genomic prediction method that integrates the gating multilayer perceptron and linear attention mechanisms[J].Molecular Plant, 2025, 18(3): 527-549.

[78] Li Y, Ren S, Li J, et al. MeNet: a mixed-effect deep neural network for multi-environment genomic prediction of agronomic traits[J].Plant Communications, 2026, 7(3): 101620.

[79] You F, Zheng C, Daniel J J Z, et al. MultiGS: a comprehensive and user-friendly genomic prediction platform integrating statistical, machine learning, and deep learning models for breeders [J/OL].Preprint at Research Square, 2026 [2026−05−01].https://doi.org/10.64898/2026.01.02.697306.

[80] Swain M K, Kamila N K, Jena L, et al. Hybrid deep learning framework for accurate classification of high dimensional genomic data[J].Scientific Reports, 2026, 16(1): 5919.

[81] Novielli P, Romano D, Pavan S, et al. Explainable artificial intelligence for genotype-to-phenotype prediction in plant breeding: a case study with a dataset from an almond germplasm collection[J].Frontiers in Plant Science, 2024, 15: 1434229.

[82] Montesinos-López O A, Solís-Covarrubias A E, Hernández-Suarez C M, et al. A transfer learning method for enhanced genomic prediction[J].Discover Plants, 2025, 2(1): 278.

[83] Villanueva B, Kennedy B W. Index versus tandem selection after repeated generations of selection[J].Theoretical and Applied Genetics, 1993, 85(6−7): 706-712.

[84] Yasuda Y, Iki T, Takashima Y, et al. Inheritance of growth ring components and the possibility of early selection for higher wood density in japanese cedar (cryptomeria japonica D. Don)[J].Annals of Forest Science, 2024, 81(1): 5.

[85] Moeinizade S, Kusmec A, Hu G, et al. Multi-trait genomic selection methods for crop improvement[J].Genetics, 2020, 215(4): 931-945.

[86] Hazel L N. The genetic basis for constructing selection indexes[J].Genetics, 1943, 28(6): 476-490.

[87] Cerón-Rojas J J, Crossa J. The statistical theory of linear selection indices from phenotypic to genomic selection[J].Crop Science, 2022, 62(2): 537-563.

[88] Sinha D, Maurya A K, Abdi G, et al. Integrated genomic selection for accelerating breeding programs of climate-smart cereals[J].Genes, 2023, 14(7): 1484.

[89] Ceron-Rojas J J, Crossa J, Arief V N, et al. A genomic selection index applied to simulated and real data[J].G3: Genes, Genomes, Genetics, 2015, 5(10): 2155-2164.

[90] Duarte D, Jurcic E J, Dutour J, et al. Genomic selection in forest trees comes to life: unraveling its potential in an advanced four-generation eucalyptus grandis population[J].Frontiers in Plant Science, 2024, 15: 1462285.

[91] Jesús Cerón-Rojas J, Montesinos-López Ó A, Montesinos-López A, et al. Nonlinear genomic selection index accelerates multi-trait crop improvement[J].Nature Communications, 2026, 17(1): 1991.

[92] Castro I, Salas-González R, Fidalgo B, et al. Optimising forest management using multi-objective genetic algorithms[J].Sustainability, 2024, 16(23): 10655.

[93] Vijai P. A hybrid multi-objective optimization approach with NSGA-II for feature selection[J].Decision Analytics Journal, 2025, 14: 100550.

[94] Hamilton M G. Optimal contribution selection in highly fecund species with overlapping generations[J].Journal of Heredity, 2020, 111(7): 646-651.

[95] El-Kassaby Y A, Cappa E P, Chen C, et al. Efficient genomics-based ‘end-to-end’ selective tree breeding framework[J].Heredity, 2024, 132(2): 98-105.

[96] Araujo M J, Bush D, Tambarussi E V. Quantifying genetic and genotypic gain gaps in eucalyptus: the hidden cost of ignoring inbreeding and dominance[J].Heredity, 2025, 134(9): 542-557.

[97] Gebregiwergis G T, Sørensen A C, Henryon M, et al. Controlling coancestry and thereby future inbreeding by optimum-contribution selection using alternative genomic-relationship matrices[J].Frontiers in Genetics, 2020, 11: 345.

[98] Thistlethwaite F R, Gamal El-Dien O, Ratcliffe B, et al. Linkage disequilibrium vs. pedigree: genomic selection prediction accuracy in conifer species[J].PLoS One, 2020, 15(6): e0232201.

基金资助

中国林科院中央级公益性科研院所基本科研业务费专项(CAFYBB2022ZA002)

国家自然科学基金项目(32572095)