关键词: 作物, 知识图谱, 育种知识发现, 优异多效基因

Abstract:

As the cornerstone of ensuring national food security and the effective supply of important agricultural products, the seed industry has always been the direction of breeders' efforts to cultivate new crop varieties with the aggregation of a variety of excellent traits. Therefore, the excavation of pleiotropic genes that regulate multiple excellent traits such as drought resistance and disease resistance will effectively contribute to the scientific research of crop breeding. At present, with the accelerated application of information technology in the field of crop breeding, the multi-dimensional scientific data related to crop breeding has increased exponentially. These semi-structured and structured scientific data are distributed in scientific databases in different fields, and there is a lack of cross-species and multi-dimensional scientific data correlation and fusion datasets, which hinders the migration and reuse of existing crop breeding knowledge and maximizes the value of crop breeding scientific data, which brings challenges to the discovery of crop trait regulation gene knowledge. Based on the reliability, practicability, and ease of use of the data, PubMed literature database, Phytozome, Ensembl plants, UniProt, RGAP, STRING, Pfam, KEGG and GO were selected as the data acquisition sources, and the entities and relationships of scientific data in different data formats were extracted by multi-path knowledge extraction. It is mainly oriented to the mapping knowledge extraction of structured data; For XML semi-structured data, knowledge extraction based on Kettle data analysis is adopted. For FASTA semi-structured data, knowledge extraction based on BLAST model is adopted. For Text unstructured data, knowledge extraction based on large language models is adopted. On the basis of the above entity and relationship extraction, the association and integration of multi-source crop breeding knowledge were further realized based on the entity mapping and specific attribute association. Finally, a knowledge graph dataset of crop trait regulatory genes was formed, which was stored as structured data in.csv format. The dataset consists of 13 entity datasets and 14 semantic relationship datasets. In order to verify the validity of the dataset, the Neo4j graph database was used for dataset storage. Finally, a knowledge graph of crop trait regulatory genes covering 130,000 nodes and 550,000 semantic relationships was formed, which could effectively support the association retrieval of cross-species gene knowledge. The knowledge graph dataset of crop trait regulatory genes has provided a key semantic model and an important data basis for the discovery of crop breeding knowledge such as excellent pleiotropic gene discovery, cross-species gene function prediction and pathway gene network potential discovery. Based on this dataset, relevant scientific research and production units can construct a knowledge base of crop trait regulatory genes, which provides a key knowledge resource base for the construction of a crop breeding knowledge discovery service platform.

Data summary：

Key words: crops, knowledge graph, crop breeding knowledge discovery, elite polyphenotype genes