According to Xinhua News Agency, on August 16, at the 32nd National Symposium on Soybean Research and Production held in Nanning, Guangxi, Professor Wang Xiaobo of Anhui Agricultural University unveiled “Fengshu” 2.0, a vertical large model for the soybean sector, on behalf of the development team. It enables multimodal data fusion, reduces response bias from individual models, and helps advance data-assisted soybean breeding.
The report noted that soybeans are an important multipurpose crop in China, used for food, oil, and animal feed. In February this year, Anhui Agricultural University and the Institute of Crop Sciences at the Chinese Academy of Agricultural Sciences jointly completed the development of the “Fengshu” large model. The model is a generative AI breeding platform covering the entire soybean life cycle. The two institutions recently upgraded it to a new iteration.
“Fengshu” 2.0 integrates soybean germplasm resources, genomes and coding sequences, proteins, transcripts, gene expression data, disease samples, breeding trial records, and Chinese- and foreign-language literature, creating a structured knowledge system for soybean research. Compared with the initial version, “Fengshu” 2.0 adds a collaborative multi-model analysis mechanism.
According to the development team, “Fengshu” 2.0 incorporates interfaces for several general-purpose large models. These models can each answer the same specialized soybean-related question, after which the “Fengshu” 2.0 summarization module combines domain knowledge, private data, and structured comparison rules to comprehensively analyze the consensus, differences, and completeness of evidence across their responses. It then produces clearer and more reliable expert conclusions, helping improve the consistency of soybean knowledge services.
Reporters learned that “Fengshu” 2.0 can currently assist with disease diagnosis, breeding parent selection, virtual cross design, phenotypic data analysis, marker-assisted breeding, and candidate gene function analysis, with a focus on important agronomic traits in soybeans.
IT Home found that “Fengshu” 2.0 integrates six major functional modules: Soybean Encyclopedia, Soybean Molecular Data, Soybean Literature, Soybean Diseases, Soybean Phenotypes, and Soybean Breeding. Together, they form a multidimensional data system covering knowledge graphs and literature, genomic and multi-omics data, phenomic data, and practical breeding data. The model incorporates more than 10 million Chinese characters of specialized text, over 10,000 research papers and patents, and a soybean-domain knowledge graph containing 20,000 entities and 100,000 relationships. It also integrates resequencing data from 8,000 genomes, approximately 40,000 soybean germplasm resources, and phenotypic data from more than 3,000 varieties.