A first variety of edible bean that was selected with the help of artificial intelligence should be available in the next couple of years.
Mohsen Yoosefzadeh Najafabadi, the edible bean breeder at the University of Guelph, said that since he’s been on staff, the bean breeding program is evolving to use a mix of traditional breeding with computational biology.
Part of that is the use of computers trained to make the most use of datasets around genetics and edible bean (dry bean) breeding.
“Based on what we saw, what we’ve seen so far in the field, it looks great,” he said about the first variety developed using AI assistance at the recent Ontario Bean Growers’ research day at the Huron Research Station in Centralia.
“We have field trials, genetic information, phenotyping, drone imaging, seed quality, weather and soil data right at our fingertips,” Yoosefzadeh Najafabadi says. “Using all this data, we can go from tens of thousands of suggested crosses to just a few hundred.”
The advantage is being able to make use of years and thousands of pages of data, only some of which can be managed by humans.
“How about letting AI better analyze our data and tell us which cross with the line would be better? And then we can sort it out based on our own knowledge,” he says.
Humans still cross the plant breeding lines, without machine intervention. It’s the ability to prioritize the crosses that’s powerful, says Yoosefzadeh Najafabadi.
The AI model is called BeanGPT, and it has been trained using years of data on dry bean breeding and agronomy.
BeanGPT also pulls in global information that researchers might not otherwise know.
Yoosefzadeh Najafabadi is working to develop anthracnose resistance, which became a priority after anthracnose challenges in 2024 in Ontario.
He used this example to explain the new computational biology approach applied in his lab.
A graduate student looked at the genetic side of anthracnose resistance and they developed 200 genetic lines of beans, which were then grown out and exposed to anthracnose.
The genetic lines were characterized based on their exposure to anthracnose, and then genomically tested to find the markers for resistance.
“We found three important markers, and I’ve used BeanGPT to dig into these markers to see what are they responsible for and how close they are to the current markers that we have.”
Two of the markers were already known, but one, on chromosome 11, can also be used as a marker. The researchers will now bring forward varieties with the three markers for resistance.
New ways to assess disease resistance
Another of Yoosefzadeh Najafabadi’s graduate students has looked at the ability to use hyperspectral imagery to see if a genetic line is susceptible to anthracnose in a growth chamber.
It so far looks successful.
Yoosefzadeh Najafabadi’s lab is also screening for soybean cyst nematode, and it has used technology to make that easier. More graduate students created an app which will count the cysts on bean roots, a job that is tedious and relies on human observations being accurate.
Hyperspectral imaging is also valuable for screening for SCN prevalence as the leaves of bean plants affected by SCN turn a brighter green colour before they start to turn yellow.
Yoosefzadeh Najafabadi expects to be able to accelerate the pace of bean breeding using technology, but he’s not throwing away traditional bean breeding knowledge.
“It’s not actually like having a transition from traditional to the modern, but merging traditional breeding with the computational powers to better understand what happens in the field.”
The post Mixing traditional breeding and computational biology in plant breeding appeared first on Farmtario.
