By Huang Xiaohui, People’s Daily
“Even field workers now use AI for crop breeding,” remarked Huang Fugang, a rice molecular breeder at Guangxi University, during recent acknowledgments to the Fengdeng large model’s R&D team.
Released in April 2024 by a coalition of researchers from the Shanghai Artificial Intelligence (AI) Laboratory, Yazhouwan National Laboratory, China Agricultural University, and other research institutions, the Fengdeng model laid the foundation for Fengdeng Gene Scientist – an AI-driven biological breeding launched in July 2024. This tool helps scientists explore and validate unknown gene functions.
Crop breeding hinges on designing agronomic traits through precise genome engineering. Despite heavy global investment since rice genome sequencing concluded in 2005, decoding gene functions progressed slowly. Traditional breeding remains reliant on expert intuition, with hypotheses, experiments, and validation spanning years and yielding limited success.
“It’s like deciphering a book of cryptic symbols – we decode few, and efficiency stays low,” explained Yang Fan, Yazhouwan National Laboratory researcher and Fengdeng project co-lead. He emphasized that historical breeding data, crop sequences, field conditions, and cultivation practices critically determine outcomes, adding: “With extreme weather intensifying, field dynamics shift rapidly. Relying solely on human experience makes successful breeding increasingly difficult.”
The team trained the AI agent with massive datasets, enabling it to identify gene-trait relationships, predict gene-trait associations, and design and simulate breeding experiments.
Since the global release of the Fengdeng large model’s rice-specific version in May this year, it has been adopted by leading institutions such as the International Rice Research Institute and Indian Institute of Rice Research.
According to Yang, this technology enables researchers to rapidly analyze gene functions and precisely combine superior alleles. This facilitates customized breeding for both traditional objectives (high yield, disease resistance, stress tolerance) and emerging demands (enhanced nutrition, improved flavor).
“This transcends merely teaching AI breeding knowledge – we’re training it to make scientific discoveries,” explained Dong Nanqing, fellow project co-lead and researcher at the Shanghai AI Laboratory. “The system learns to interpret breeder needs, identify relevant genes, design experiments, verify outcomes, and self-correct.”
Through continuous learning, the Fengdeng Gene Scientist has developed independent research capabilities. It simulates expert reasoning and automates the entire scientific workflow from hypothesis generation through experimental design to results analysis. In validation trials, the system supported all key research decisions (excluding physical experiments) for dozens of previously unstudied rice and maize genes.
In rice research, Fengdeng large model uncovered new functions of several genes, some regulate plant hormones that determine height, while others are tied to photosynthetic efficiency. In maize research, it accurately predicted candidate genes associated with traits like plant height and ear placement, which were later confirmed by field experiments.
The Fengdeng Gene Scientist represents just the initial phase. The research team plans to integrate additional crop, environmental, and breeding data to evolve the system into a comprehensive smart breeding platform covering all species and full breeding cycles.
