基于计算文献的大豆耦合性状知识发现研究

doi:10.13560/j.cnki.biotech.bull.1985.2025-0082

摘要/Abstract

摘要：

目的提出一种基于计算文献的自动化大豆耦合性状发现模型，以精准识别田间试验前的大豆耦合性状并解析其遗传网络，弥补传统实验室方法研究在该领域的不足，为大豆育种研究提供更高效的知识发现途径。方法首先，依据权威的领域本体构建大豆语料的标注策略，以语义三元组的形式表示大豆耦合性状知识；其次，构建基于领域词典的语义三元组抽取模型，利用AdaPU（adapted positive-unlabeled learning）算法和R-BERT（pre-trained transformer encoder for relation extractio）算法自动化抽取大豆育种相关文献中的知识实体及其调控关系，得到大豆性状的遗传调控网络；最后挖掘网络中存在的耦合性状连通子网和性状节点间的可达路径，利用文献回溯的方法进行验证并进行遗传机制分析。结果该研究所提模型的准确率为79.41%，召回率为88.52%，F1 score为83.72%，获得唯一的大豆性状知识三元组776个，包含33个基因概念、119个蛋白质概念和96个性状概念，其中478个为“相关关系”，264个为“上调关系”，34个为“下调关系”。研究发现6个耦合性状连通子网，挖掘到139条性状耦合路径，揭示了大豆不同性状间的复杂关联及其潜在的分子机制。结论证实基于大规模文献进行性状知识发现的可行性，通过自动化模型挖掘并验证了大豆耦合性状及其遗传调控网络，为大豆育种领域的研究人员提供潜在的多效基因和性状关联信息，有效支撑育种实验设计和假设生成。

关键词: 知识抽取, 知识发现, 语义三元组, 大豆育种, 耦合性状

Abstract:

Objective This study aims to develop an automated soybean trait discovery model using computational literature. The goal is to accurately identify soybean coupling traits before field trials and analyze their genetic networks. This approach addresses the limitations of traditional laboratory methods and offers a more efficient way to discover knowledge for soybean breeding research. Method Firstly, the annotation strategy of soybean corpus was constructed according to the authoritative domain ontology, and the soybean coupling trait knowledge was represented in the form of Subject-Predication-Object (SPO) semantic triples. Secondly, a semantic triplet extraction model was constructed based on the domain dictionary. Adapted positive-unlabeled learning (AdaPU) algorithm and R-BERT (pre-trained Transformer encoder for relation extraction) algorithm were used to automatically extract the knowledge entities and their regulatory relationships in soybean breeding literature, and the genetic regulatory networks of soybean traits were obtained. Finally, the coupling trait connected subnetworks and the reachable paths between trait nodes in the network were mined, and the literature review method was used to verify the results and analyze the genetic mechanism. Result Experimental results show that the knowledge extraction model achieved an accuracy of 79.41%, a recall rate of 88.52%, and an F1 score of 83.72%. A total of 776 unique soybean trait knowledge triples were identified, encompassing 33 gene concepts, 119 protein concepts, and 96 trait concepts. Among these, 478 triples represented “associated with” relationships, 264 “up-regulation” relationships, and 34 “down-regulation” relationships. Within the soybean trait knowledge network, six coupling traits connected subgraphs were discovered, and 139 trait coupling paths within the largest connected subnetwork. Conclusion This study identified the feasibility of trait knowledge discovery based on large-scale literature. By deeply mining knowledge units, it uncovers the underlying coupling traits and their associated molecular mechanisms, providing plant breeding researchers with potential pleiotropic genes and coupling traits for experimental design, thereby enhancing the efficiency of hypothesis generation in scientific research. The structured trait knowledge generated by this study contributes to the development of knowledge graphs in the field of soybean breeding and serves as a reliable knowledge foundation for domain-specific large language models, facilitating the development and application of AI agent.

Key words: knowledge extraction, knowledge discovery, SPO, soybean breeding, coupling traits

关陟昊, 单治易, 熊赫, 赵瑞雪. 基于计算文献的大豆耦合性状知识发现研究[J]. 生物技术通报, 2025, 41(9): 345-356.

GUAN Zhi-hao, SHAN Zhi-yi, XIONG He, ZHAO Rui-xue. Computational Literature-based Knowledge Discovery for Soybean Coupling Traits[J]. Biotechnology Bulletin, 2025, 41(9): 345-356.

图/表 11

参考文献 25

[1]	Tam V, Patel N, Turcotte M, et al. Benefits and limitations of genome-wide association studies [J]. Nat Rev Genet, 2019, 20(8): 467-484.
[2]	Nédellec C, Sauvion C, Bossy R, et al. TaeC: a manually annotated text dataset for trait and phenotype extraction and entity linking in wheat breeding literature [J]. PLoS One, 2024, 19(6): e0305475.
[3]	Yacoubi Ayadi N, Bernard S, Bossy R, et al. A unified approach to publish semantic annotations of agricultural documents as knowledge graphs [J]. Smart Agric Technol, 2024, 8: 100484.
[4]	Gao YJ, Zhou Q, Luo JX, et al. Crop-GPA an integrated platform of crop gene-phenotype associations [J]. NPJ Syst Biol Appl, 2024, 10(1): 15.
[5]	Lotreck S, Segura Abá K, Lehti-Shiu MD, et al. Plant science knowledge graph corpus: a gold standard entity and relation corpus for the molecular plant sciences [J]. Silico Plants, 2024, 6(1): diad021.
[6]	Lee HJ, Chung YJ, Jang S, et al. Genome-wide identification of major genes and genomic prediction using high-density and text-mined gene-based SNP panels in Hanwoo (Korean cattle) [J]. PLoS One, 2020, 15(12): e0241848.
[7]	Singh G, Papoutsoglou EA, Keijts-Lalleman F, et al. Extracting knowledge networks from plant scientific literature: potato tuber flesh color as an exemplary trait [J]. BMC Plant Biol, 2021, 21(1): 198.
[8]	Xie CJ, Gao J, Chen JJ, et al. PotatoG-DKB a potato gene-disease knowledge base mined from biological literature [J]. PeerJ, 2024, 12: e18202.
[9]	Liu YY. DKG-PIPD: a novel method about building deep knowledge graph [J]. IEEE Access, 2021, 9: 137295-137308.
[10]	Lu J, Yang WX, He L, et al. A method for extracting fine-grained knowledge of the wheat production chain [J]. Agronomy, 2024, 14(9): 1903.
[11]	Yuan WW, Yang WX, He L, et al. Research on entity and relationship extraction with small training samples for cotton pests and diseases [J]. Agriculture, 2024, 14(3): 457.
[12]	Yang K, Liu Y. Construction of knowledge graph in the field of grassland plants based on ontology database[C]//2021 International Conference on Environmental Remote Sensing and Big Data. Wuhan China: SPIE, 2021: 12129.
[13]	Lou DJ, Li F, Ge JY, et al. LncPheDB: a genome-wide lncRNAs regulated phenotypes database in plants [J]. aBIOTECH, 2022, 3(3): 169-177.
[14]	Fang C, Ma YM, Wu SW, et al. Genome-wide association studies dissect the genetic networks underlying agronomical traits in soybean [J]. Genome Biol, 2017, 18(1): 161.
[15]	Singh G, Kuzniar A, Brouwer M, et al. Linked data platform for Solanaceae species [J]. Appl Sci, 2020, 10(19): 6813.
[16]	Ivanisenko TV, Saik OV, Demenkov PS, et al. The Solanum tuberosum knowledge base: the section on molecular-genetic regulation of metabolic pathways [J]. Vestn VOGiS, 2018, 22(1): 8-17.
[17]	Brown AV, Conners SI, Huang W, et al. A new decade and new data at SoyBase, the USDA-ARS soybean genetics and genomics database [J]. Nucleic Acids Res, 2021, 49(D1): D1496-D1501.
[18]	Zhang F, Ma LD, Wang JP, et al. An MRC and adaptive positive-unlabeled learning framework for incompletely labeled named entity recognition [J]. Int J Intell Syst, 2022, 37(11): 9580-9597.
[19]	Wu SC, He YF. Enriching pre-trained language model with entity information for relation classification [C]//Proceedings of the 28th ACM International Conference on Information and Knowledge Management. Beijing China. New York: ACM, 2019: 2361-2364.
[20]	Yuan XB, Jiang XY, Zhang MZ, et al. Integrative omics analysis elucidates the genetic basis underlying seed weight and oil content in soybean [J]. Plant Cell, 2024, 36(6): 2160-2175.
[21]	Wang XY, Yu RB, Wang JJ, et al. The asymmetric expression of SAUR genes mediated by ARF7/19 promotes the gravitropism and phototropism of plant hypocotyls [J]. Cell Rep, 2020, 31(3): 107529.
[22]	Wang CX, Ji YH, Cao XS, et al. Carbon dots improve nitrogen bioavailability to promote the growth and nutritional quality of soybeans under drought stress [J]. ACS Nano, 2022, 16(8): 12415-12424.
[23]	Shi SY, Miao HY, Du XM, et al. GmSGR1, a stay-green gene in soybean (Glycine max L.), plays an important role in regulating early leaf-yellowing phenotype and plant productivity under nitrogen deprivation [J]. Acta Physiol Plant, 2016, 38(4): 97.
[24]	Zhang JP, Wang XZ, Lu YM, et al. Genome-wide scan for seed composition provides insights into soybean quality improvement and the impacts of domestication and breeding [J]. Mol Plant, 2018, 11(3): 460-472.
[25]	Hassani-Pak K, Singh A, Brandizi M, et al. KnetMiner: a comprehensive approach for supporting evidence-based gene discovery and complex trait analysis across species [J]. Plant Biotechnol J, 2021, 19(8): 1670-1678.

实体类型Entity type	实体开始位置标注 Annotation of entity start position	实体内部位置标注 Annotation of entity internal position
基因 Gene	B-GEN	I-GEN
蛋白质 Protein	B-PRO	I-PRO
性状 Trait	B-TRA	I-TRA
非实体 Non-entity	O	O

实体类型Entity type	实体开始位置标注 Annotation of entity start position	实体内部位置标注 Annotation of entity internal position
基因 Gene	B-GEN	I-GEN
蛋白质 Protein	B-PRO	I-PRO
性状 Trait	B-TRA	I-TRA
非实体 Non-entity	O	O

关系类型 Relation type	关系标注 Relation annotation
e1上调e2	UP（e1， e2）
e1下调e2	DP（e1， e2）
e1与e2相关	AW（e1， e2）

关系类型 Relation type	关系标注 Relation annotation
e1上调e2	UP（e1， e2）
e1下调e2	DP（e1， e2）
e1与e2相关	AW（e1， e2）

文本序列 Text sequence	文本标注 Text annotation
A	O
GmPP2C-1	B-GEN
Gene	O
From	O
Wild	O
Soybean	O
Helps	O
To	O
Increase	O
Seed	B-TRA
Weight	I-TRA
	UP （GmPP2C-1， seed weight）