[1]郭茂祖,周遨宇,段然.融合多实例学习与注意力机制的异构体功能预测方法[J].智能系统学报,2025,20(6):1508-1519.[doi:10.11992/tis.202410005]
GUO Maozu,ZHOU Aoyu,DUAN Ran.Isoform function prediction based on attention mechanism and multiple instance learning[J].CAAI Transactions on Intelligent Systems,2025,20(6):1508-1519.[doi:10.11992/tis.202410005]
点击复制
GUO Maozu,ZHOU Aoyu,DUAN Ran.Isoform function prediction based on attention mechanism and multiple instance learning[J].CAAI Transactions on Intelligent Systems,2025,20(6):1508-1519.[doi:10.11992/tis.202410005]
融合多实例学习与注意力机制的异构体功能预测方法
《智能系统学报》[ISSN 1673-4785/CN 23-1538/TP] 卷:
20
期数:
2025年第6期
页码:
1508-1519
栏目:
吴文俊人工智能科学技术奖论坛
出版日期:
2025-11-05
- Title:
- Isoform function prediction based on attention mechanism and multiple instance learning
- Keywords:
- gene functions; high-resolution annotation; isoform functions; graph convolutional network; gene ontology embedding; isoform interaction network; fusion network; attention-weighted; loss function
- 分类号:
- TP181
- 摘要:
- 基因功能的高分辨率注释是功能基因组学的核心任务。单个基因可变剪接产生的异构体（isoform）翻译出多种蛋白质变体，为生物体提供了功能多样性。为实现异构体功能的高分辨率注释，本文提出了一种方法LossIsoFun。引入基因本体（gene ontology，GO），并利用图卷积神经网络（graph convolutional network，GCN）保留其层次结构和语义信息，通过GO网络嵌入策略获得压缩的基因GO注释。融合异构体互作网络、共表达网络和序列相似性网络，构建异构体功能网络，并将异构体序列数据与功能网络输入GCN，获取异构体功能的低维表示。通过基因与异构体的关联关系，得到基因功能的低维表示。提出一种基于注意力权重的损失函数，通过最小化压缩的基因GO注释与基因功能低维表示之间的差异来训练模型。通过解压缩异构体的低维表示，获得异构体的高分辨率注释。在人类基准数据集上的对比实验验证了LossIsoFun的有效性。
- Abstract:
- High-resolution annotation of gene functions is essential in functional genomics. Multiple isoforms are generated from a single gene via alternative splicing, thereby producing protein variants that contribute to functional diversity. This paper introduces LossIsoFun, a framework for high-resolution isoform function annotation. First, gene ontology (GO) and a graph convolutional network (GCN) are used to preserve hierarchical and semantic structures, producing compressed GO annotations. Then, isoform interaction, coexpression, and sequence similarity networks are integrated to construct an isoform functional network. The isoform sequence data and functional network are fed into a GCN to generate low-dimensional isoform representations. By leveraging gene–isoform relationships, gene function representations are derived. A novel loss function minimizes differences between compressed GO annotations and gene function representations. Finally, isoform functions are annotated by decompressing these representations. Validation on human benchmark datasets demonstrates that LossIsoFun effectively yields isoform function annotation.
- 参考文献/References:
-
[1] PAN Qun, SHAI O, LEE L J, et al. Deep surveying of alternative splicing complexity in the human transcriptome by high-throughput sequencing[J]. Nature genetics, 2008, 40(12): 1413-1415.
[2] WANG E T, SANDBERG R, LUO Shujun, et al. Alternative isoform regulation in human tissue transcriptomes[J]. Nature, 2008, 456(7221): 470-476.
[3] CROWL S, COLEMAN M B, CHAPIV A, et al. Systematic analysis of the effects of splicing on the diversity of post-translational modifications in protein isoforms using PTM-POSE[EB/OL]. (2024-01-11)[2025-09-15].https://doi.org/10.1101/2024.01.10.575062 .
[4] SMITH L M, KELLEHER N L. Proteoforms as the next proteomics currency[J]. Science, 2018, 359(6380): 1106-1107.
[5] 曾杰. 基于深度多示例学习的可变剪接异构体相互作用预测研究[D]. 重庆: 西南大学, 2021.
ZENG Jie. Study on interaction prediction of alternative splicing isomers based on deep multi-instance learning[D]. Chongqing: Southwest University, 2021.
[6] HOWES A, ROGERSON C, BELYAEV N, et al. The FAM13A long isoform regulates cilia movement and coordination in airway mucociliary transport[J]. American journal of respiratory cell and molecular biology, 2024, 71(3): 282-293.
[7] MITTENDORF K F, DEATHERAGE C L, OHI M D, et al. Tailoring of membrane proteins by alternative splicing of pre-mRNA[J]. Biochemistry, 2012, 51(28): 5541-5556.
[8] GUO Miao, LIU Wei, SERRA S, et al. FGFR2 isoforms support epithelial-stromal interactions in thyroid cancer progression[J]. Cancer research, 2012, 72(8): 2017-2027.
[9] WANG Shiying, SUN Boyun, YUAN Jianye, et al. The different effects of VEGFA121 and VEGFA165 on regulating angiogenesis depend on phosphorylation sites of VEGFR2[J]. Inflammatory bowel diseases, 2017, 23(4): 603-616.
[10] HASSN MESRATI M, SYAFRUDDIN S E, MOHTAR M A, et al. CD44: a multifunctional mediator of cancer progression[J]. Biomolecules, 2021, 11(12): 1850.
[11] REVIL T, TOUTANT J, SHKRETA L, et al. Protein kinase C-dependent control of Bcl-x alternative splicing[J]. Molecular and cellular biology, 2007, 27(24): 8431-8441.
[12] ASHBURNER M, BALL C A, BLAKE J A, et al. Gene ontology: tool for the unification of biology[J]. Nature genetics, 2000, 25(1): 25-29
[13] ZHAO Yingwen, WANG Jun, GUO Maozu, et al. Cross-species protein function prediction with asynchronous-random walk[J]. IEEE/ACM transactions on computational biology and bioinformatics, 2019, 18(4): 1439-1450.
[14] ZHAO Yingwen, FU Guangyuan, WANG Jun, et al. Gene function prediction based on gene ontology hierarchy preserving hashing[J]. Genomics, 2019, 111(3): 334-342.
[15] YU Guoxian, WANG Keyao, FU Guangyuan, et al. NMFGO: gene function prediction via nonnegative matrix factorization with gene ontology[J]. IEEE/ACM transactions on computational biology and bioinformatics, 2020, 17(1): 238-249.
[16] CARBONNEAU M A, CHEPLYGINA V, GRANGER E, et al. Multiple instance learning: a survey of problem characteristics and applications[J]. Pattern recognition, 2018, 77: 329-353.
[17] CHEN Hao, SHAW D, ZENG Jianyang, et al. DIFFUSE: predicting isoform functions from sequences and expression profiles via deep learning[J]. Bioinformatics, 2019, 35(14): i284-i294.
[18] LI Wenyuan, KANG Shuli, LIU Chunchi, et al. High-resolution functional annotation of human transcriptome: predicting isoform functions by a novel multiple instance-based label propagation method[J]. Nucleic acids research, 2014, 42(6): e39.
[19] YU Guoxian, WANG Keyao, DOMENICONI C, et al. Isoform function prediction based on bi-random walks on a heterogeneous networkFree[J]. Bioinformatics, 2020, 36(1): 303-310.
[20] SHAW D, CHEN Hao, JIANG Tao. DeepIsoFun: a deep domain adaptation approach to predict isoform functionsFree[J]. Bioinformatics, 2018, 35(15): 2535-2544.
[21] LI Hongdong, YANG Changhuo, ZHANG Zhimin, et al. IsoResolve: predicting splice isoform functions by integrating gene and isoform-level features with domain adaptation[J]. Bioinformatics, 2021, 37(4): 522-530.
[22] QIU Sichao, YU Guoxian, LU Xudong, et al. Isoform function prediction by gene ontology embedding[J]. Bioinformatics, 2022, 38(19): 4581-4588.
[23] YU Guoxian, ZHOU Guangjie, ZHANG Xiangliang, et al. DMIL-IsoFun: predicting isoform function using deep multi-instance learning[J]. Bioinformatics, 2021, 37(24): 4818-4825.
[24] 王可尧. 基于RNA-seq数据的可变剪接异构体功能预测方法研究[D]. 重庆: 西南大学, 2019.
WANG Keyao. Study on function prediction method of alternative splicing isomers based on RNA-seq data[D]. Chongqing: Southwest University, 2019.
[25] SU Yaqi, YU Zhejian, JIN Siqian, et al. Comprehensive assessment of mRNA isoform detection methods for long-read sequencing data[J]. Nature communications, 2024, 15(1): 3972.
[26] WANG Keyao, WANG Jun, DOMENICONI C, et al. Differentiating isoform functions with collaborative matrix factorization[J]. Bioinformatics, 2020, 36(6): 1864-1871.
[27] KIPF T N, WELLING M. Semi-supervised classification with graph convolutional networks[EB/OL]. (2016-09-09)[2025-09-15]. https://arxiv.org/abs/1609.02907.
[28] 张硕. 基于图神经网络的剪接异构体功能预测方法研究[D]. 长沙: 中南大学, 2022.
ZHANG Shuo. Study on function prediction method of splicing isomers based on graph neural network[D]. Changsha: Central South University, 2022.
[29] GAO Tianyu, YAO Xingcheng, CHEN Danqi. SimCSE: simple contrastive learning of sentence embeddings[EB/OL]. (2021-04-18)[2025-09-15]. https://arxiv.org/abs/2104.08821.
[30] ZHAO Yingwen, WANG Jun, CHEN Jian, et al. A literature review of gene function prediction by modeling gene ontology[J]. Frontiers in genetics, 2020, 11: 400.
[31] LIN Dekang. An information-theoretic definition of similarity[C]//Proceedings of the Fifteenth International Conference on Machine Learning. Madison: Morgan Kaufmann Publishers Inc. , 1998: 296-304.
[32] LUO Tingjin, ZHANG Weizhong, QIU Shuang, et al. Functional annotation of human protein coding isoforms via non-convex multi-instance learning[C]//Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. Halifax: ACM, 2017: 345-354.
[33] ALTSCHUL S F, GISH W, MILLER W, et al. Basic local alignment search tool[J]. Journal of molecular biology, 1990, 215(3): 403-410.
[34] RAPPOPORT N, SHAMIR R. NEMO: cancer subtyping by integration of partial multi-omic data[J]. Bioinformatics, 2019, 35(18): 3348-3356.
[35] WANG Bo, MEZLINI A M, DEMIR F, et al. Similarity network fusion for aggregating data types on a genomic scale[J]. Nature methods, 2014, 11(3): 333-337.
[36] 赵璐, 袁立明, 郝琨. 多示例学习算法综述[J]. 计算机科学, 2022, 49(S1): 93-99.
ZHAO Lu, YUAN Liming, HAO Kun. A survey of multi-instance learning algorithms[J]. Computer science, 2022, 49(S1): 93-99.
[37] EKSI R, LI Hongdong, MENON R, et al. Systematically differentiating functions for alternatively spliced isoforms through integrating RNA-seq data[J]. PLoS comput biol, 2013, 9(11): e1003314.
[38] ZHANG Shijia, LIU Huili, YUAN Li, et al. Recognition of CCA1 alternative protein isoforms during temperature acclimation[J]. Plant cell reports, 2021, 40(2): 421-432.
[39] LANGFELDER P, HORVATH S. WGCNA: an R package for weighted correlation network analysis[J]. BMC bioinformatics, 2008, 9: 559.
[40] HUERTA-CEPAS J, SZKLARCZYK D, HELLER D, et al. eggNOG 5.0: a hierarchical, functionally and phylogenetically annotated orthology resource based on 5090 organisms and 2502 viruses[J]. Nucleic acids research, 2019, 47(D1): D309-D314.
[41] CONSORTIUM U. UniProt: the universal protein knowledgebase in 2021[J]. Nucleic acids research, 2021, 49(D1): D480-D489.
备注/Memo
收稿日期:2024-10-9。
基金项目:国家自然科学基金重点项目（62031003）；国家自然科学基金青年基金项目（62301021）.
作者简介:郭茂祖，教授，博士生导师，北京建筑大学智能科学与技术学院院长，中国人工智能学会机器学习专委会常委、中国建筑学会计算性设计学术委员会常委，主要研究方向为机器学习、计算生物学。获吴文俊人工智能自然科学奖二等奖。发表学术论文100余篇。 E-mail：guomaozu@bucea.edu.cn。;周遨宇，硕士研究生，主要研究方向为深度学习和生物信息学。E-mail：18336331205@163.com。;段然，讲师，主要研究方向为生物信息学、网络科学、数据挖掘、机器学习。主持国家自然科学基金青年项目1项。发表学术论文8篇。E-mail：duanran@bucea.edu.cn。
通讯作者:段然. E-mail：duanran@bucea.edu.cn
更新日期/Last Update:
1900-01-01