Artificial intelligence Photo: VCG

A research team from Anhui Agricultural University has unveiled "Fengshu" 2.0, a vertical large model for the soybean sector, at the 32nd National Soybean Research and Production Symposium held in Nanning, South China's Guangxi Zhuang Autonomous Region. 

The upgraded model achieves multimodal data fusion, reduces response deviations of single models, and helps advance data-assisted soybean breeding, the Xinhua News Agency reported on Monday.

Soybean is a strategically important crop in China, serving simultaneously as a source of grain, edible oil, and feed. In February, Anhui Agricultural University and the Institute of Crop Sciences of the Chinese Academy of Agricultural Sciences jointly built the original large model - a generative AI breeding platform covering the full life cycle of soybean. The two institutions recently completed an iterative upgrade of the system.

This model integrates soybean germplasm resources, genomes and coding sequences, proteins, transcripts, gene-expression profiles, disease samples, breeding trial records, and Chinese and foreign literature to form a structured knowledge system oriented to soybean science. Compared with the initial version, "Fengshu" 2.0 adds a multi-model collaborative analysis mechanism.

According to the R&D team, the new model embeds interfaces of several general-purpose large models. 

These models can each answer the same professional soybean question, after which the summary module of "Fengshu" 2.0 - using domain knowledge, private data, and structured comparison rules - comprehensively analyzes the consensus, differences, and evidentiary completeness of the answers, producing clearer and more reliable professional conclusions. This helps improve the stability of soybean-specific knowledge services.

The current version can assist, around key soybean agronomic traits, in disease diagnosis, parental line screening, virtual cross combination design, phenotypic data parsing, molecular marker-assisted breeding, and candidate gene function analysis.

"Compared with traditional empirical breeding and trial-based breeding, the new model drives soybean breeding toward data-assisted breeding and precision decision-making," Wang Xiaobo, professor with the Anhui Agricultural University said. 

"It helps improve the efficiency of scientific data processing, cut repetitive work, and optimize breeding schemes," he said. The model will further incorporate climate, soil, and planting patterns of major soybean-producing regions in China to support improved variety propagation, scaled cultivation, scientific analysis, and agro-technical extension.

The team will continue to push forward field trials, accumulation of professional data, and algorithm iteration, strengthen verification between model outputs and real breeding experiment, and expand the platform's application in germplasm mining, trait analysis, variety selection, and production management, so as to facilitate the cultivation of high-yield, high-quality, and stress-resistant soybean varieties.