For decades, the agricultural industry has wrestled with a difficult balancing act. Farmers must protect valuable crop yields from destructive pests to ensure global food security, yet standard chemical treatments frequently threaten vital pollinators like honey bees. These insects are essential not only for the health of local ecosystems but also for the commercial production of a vast array of fruits, vegetables, and nuts. Finding a method to shield crops without harming the very creatures responsible for pollinating a significant portion of the human diet has long been a holy grail for agricultural scientists.
Biologist Anandasankar Ray, a prominent expert on insect olfactory behavior at UC Riverside, set out to tackle the central problem standing in the way of this protective strategy: the extreme complexity of the honey bee olfactory system. Unlike many other insects, honey bees possess an exceptionally sophisticated sense of smell. With more than 200 distinct odor receptors capable of detecting a vast, dizzying array of volatile compounds, identifying specific scents that would reliably repel bees rather than attracting them proved to be an immense scientific challenge.
"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," Professor Ray explained. "Our goal was to find a way to use scent as a deterrent—safely and effectively."
To overcome the physiological hurdles presented by the honey bee’s acute sense of smell, Ray’s research team joined forces with fellow UC Riverside entomologist Boris Baer and members of his laboratory, who specialize in honey bee biology and behavior. Together, the interdisciplinary team turned to modern technology, developing an advanced machine-learning model designed to predict how insects would react to various chemical structures.
The machine-learning model was meticulously trained using a combination of existing chemical structure data for thousands of odorants and previously recorded behavioral responses of bees. To further refine the system and ensure its predictive accuracy, the research team supplemented this data with new behavioral observations gathered in their own laboratories using both honey bees and Drosophila, commonly known as fruit flies. This integration of empirical biological data with computational power allowed the researchers to forecast insect olfactory responses with a high degree of precision.
Once the computational system was fully optimized, it was deployed to screen a massive digital library containing more than 50 million chemical compounds. Through this high-throughput computational filtering process, the algorithm successfully narrowed down the immense list, ultimately identifying approximately 130 distinct chemical candidates predicted to possess strong potential as safe and effective bee repellents.
Following the computational screening phase, the researchers moved from the digital realm to the laboratory workbench, putting the top-performing chemical candidates to rigorous physical tests. When exposed to these selected compounds, live honey bees exhibited clear, consistent avoidance behaviors. These empirical reactions aligned remarkably well with the predictions generated by the machine-learning model, validating the accuracy of the computational approach.
Encouraged by the laboratory results, the team advanced to field experiments involving freely foraging bees. These real-world tests confirmed that all seven of the specific compounds evaluated in the field reliably repelled bees from honey combs without causing any harm to the insects’ health or well-being.
"By keeping bees away from harmful pesticides, we can potentially reduce their risk of exposure without compromising the protection of crops," Professor Ray noted in a media release detailing the findings.
The unintended exposure of foraging bees to agricultural chemicals is widely considered a major contributing factor to overall population declines and the broader phenomenon of colony collapse disorder. While regulatory bodies and agricultural scientists have worked for years to ban or heavily restrict the most acutely toxic pesticides, Ray emphasized that significant opportunities remain to further lower environmental risks. By integrating effective, non-toxic repellents directly into pesticide management strategies, farmers can minimize direct contact between pollinators and agricultural chemicals.
Beyond the immediate agricultural applications in crop protection, the newly identified bee-repelling compounds hold promise for a variety of other settings where human-wildlife proximity can lead to conflict.
"In certain public environments—hospitals, office buildings, and residential areas, for example—reducing the formation of beehives can help avoid human-bee conflicts," Ray said. He also pointed out commercial practices that intentionally seek to avoid pollination altogether, such as the cultivation of specific seedless fruit varieties where pollination is unnecessary or even economically undesirable.
The research findings have been published in the scientific journal eLife. Furthermore, Professor Ray is listed as a co-inventor on a formal patent application covering the newly identified chemical compounds, representing a tangible step toward translating academic discovery into commercial, industry-ready products. The ultimate vision is the development of specialized, bee-friendly pesticide formulations that safeguard pollinators while simultaneously meeting the rigorous demands of modern, large-scale agriculture.
"Protecting pollinators doesn’t have to come at the expense of food security," Ray said, reflecting on the broader implications of the project. "We believe our work will help guide further exploration of machine learning-guided solutions in environmental protection and sustainable farming practices."
The success of the project also challenges conventional assumptions regarding the data requirements necessary for effective artificial intelligence applications in biology. While many advanced machine-learning initiatives rely on massive, exhaustive datasets to yield reliable results, Ray noted that olfaction presents a different paradigm.
"It is generally thought you need abundant data to do any kind of machine learning, but that’s not true for olfaction," Ray said. "This is a powerful demonstration of how machine learning can help solve real-world ecological problems."
