Machine Learning on the Case to Help Protect Important Pollinators

Researchers at the University of California, Riverside, have developed a machine learning-based method to identify chemical compounds that can safely repel honey bees from pesticide-treated crops. The research, funded by a grant from the California Research Alliance by BASF, offers a potential solution to one of agriculture’s most urgent ecological challenges: the ongoing decline in honey bee populations, driven in part by pesticide exposure.

The interdisciplinary team led by Anandasankar Ray, a Professor of molecular, cell and systems biology and an expert on insect olfactory behavior, tackled a central obstacle in altering behavior of pollinators to protect them: the complexity of the honey bee olfactory system. With more than 200 odor receptors capable of detecting a vast array of volatile compounds, finding scents that repel bees was a major challenge — until now.

“Bees rely heavily on their sense of smell to forage, but that sensitivity makes it tough to find odors that push them away instead of drawing them in,” Ray says. “Our goal was to flip that script and find a way to use scent as a deterrent — safely and effectively.”

Researchers created a machine-learning model trained on both the chemical structures of odorants and previously recorded behavioral responses of bees. The model was refined with new behavioral data the team obtained in the lab from honey bees and Drosophila (fruit flies), allowing for a more accurate prediction of insect olfactory responses. Once optimized, the system screened more than 50 million compounds, ultimately identifying about 130 that were predicted to have strong potential as bee repellents.

“It is generally thought you need abundant data to do any kind of machine learning, but that’s not true for olfaction,” Ray says. “You simply need good-quality data and iterative improvement steps.”

The team reports in the journal eLife how they put the top-performing candidates to the test. In the lab, honey bees exhibited clear avoidance behaviors when exposed to these candidates, aligning closely with the model’s predictions. Subsequent field experiments with freely foraging bees confirmed that all seven compounds tested reliably repelled bees from honey combs without harming them.

“This is a powerful demonstration of how machine learning can help solve real-world ecological problems,” Ray adds. “By keeping bees away from harmful pesticides, we can potentially reduce their risk of exposure without compromising the protection of crops.”

For more, continue reading at news.ucr.edu.