Initiative Result:

Artificial intelligence helps unearth genebanks’ hidden gems

IRRI is screening 60,000 rice samples for resilience to flooding in one year. Three times more than were screened in the previous 52 years.

Finding the most valuable samples in CGIAR genebanks has – until now – been a slow, expensive process. IRRI is changing this by using artificial intelligence (AI) to accelerate the screening of its 132,000 rice samples. This will help identify varieties that can tolerate stresses like drought or flooding. It is part of a wider program to curate richer data and insight to help breeders, farmers and researchers make full use of CGIAR’s unique collections.

CGIAR genebanks play a vital role in conserving and distributing plant material. But they do much more than that. They also carry out extensive research to help users find the varieties or traits they need. Traditional methods of evaluating collections are laborious and costly. For example, technicians visually assess seeds against scales in a manual for 80 or more features like “ligule shape” and “leaf blade color intensity”. When carrying out field assessments, researchers plant samples one-by-one, carefully marking them with physical labels.

As a result, a lot of the material in genebanks remains untapped as very little is known about their traits and characteristics. In the case of IRRI’s genebank, only around 5 percent of the collection has been used in breeding at IRRI. To address this, IRRI is now using technology like AI in the screening process. This makes it possible to generate information about the varieties more quickly, notably about their tolerance of climatic stresses like flooding and drought.

Partly financed by a USD $2 million grant from, IRRI has trained an AI tool to recognize rice samples using images of the seeds. The machine can do this much faster and more reliably than humans. This means that rather than planting the samples individually, taking care to keep them labelled and separated, researchers can sow them in bulk.

To test for flooding tolerance, they then submerge the plants in water and see which ones survive. For drought resilience, they use drones to measure the plants’ ability to generate biomass when deprived of water. In both cases, the AI tool makes it easy to work back to the original variety from the seeds of the successful plants.

Using this approach, IRRI is now screening 60,000 of its rice samples for tolerance to flooding in just one year. That would have been unthinkable in the past. Before this project started, only about 20,000 samples had been screened since the genebank opened in 1972. As well as speeding up the process, it will dramatically reduce the costs. IRRI estimates that it should be possible to screen its entire collection for around 6 percent of the cost using the traditional approach.

This type of technique could lead to a step change in how intensively genebank resources are used. If collections can be screened in bulk for some traits, then it could lead to a considerable increase in germplasm distributions from the banks. The Google-funded project contributed to 2023 being a record year for distributions (209,711 samples). When other Centers adopt similar technology, we may well see new records broken.

Using AI for screening is part of a wider program to develop a digital ecosystem for users of CGIAR genebanks. Different users have different needs. Breeders may need detailed genomic and trait information. Farmers may need policy guidelines or information on the environments in which varieties can be grown. The Initiative is therefore developing a range of digital tools so that users have more information to find the precise resources they need.

Multispectral imaging, now in use in six genebanks, is another example of a high throughput imaging process that can be used to identify and describe seeds, other plant materials, and even pathogens. Much faster and more reliable than the human eye and cheaper than DNA barcoding, such imaging methods are changing how genebanks are managed and used.

The Initiative is sharing experience and expertise in using these and other technologies with national genebanks in low- and middle-income countries, leading to more widespread and targeted use of crop diversity around the world.

Technologies like AI help us find the hidden treasures in genebanks much faster. This unlocks the genebanks’ full potential, giving more options to farmers and breeders who need to grow and develop crops that can thrive in a changing climate.

Dr. Venuprasad Ramaiah, Head, Fit-for-Future Genetic Resources, IRRI

Header photo: Rice accessions for screening forflood tolerance. IRRI.

CGIAR Centers

Primary: IRRI – International Rice Research Institute


Japan International Research Center for Agricultural Sciences (JIRCAS); Okayama University (OU)