Machine learning identifies chemicals to repel honey bees from pesticides

Researchers at the University of California, Riverside, have successfully developed a machine learning-driven methodology to identify chemical compounds that safely deter honey bees from pesticide-treated crops. Published in the journal eLife, the study addresses a critical ecological challenge: the declining global honey bee population, significantly impacted by unintended pesticide exposure. Led by Professor Anandasankar Ray, an expert in insect olfactory behavior, and developed in collaboration with entomologist Professor Boris Baer, the interdisciplinary team confronted the inherent complexity of the bee olfactory system, which comprises over 200 odor receptors capable of detecting a vast spectrum of volatile compounds. To overcome this challenge, the team engineered a predictive machine learning model trained on the chemical structures of odorants alongside behavioral response data from honey bees and Drosophila. Ray noted that olfactory modeling does not require massive datasets, only high-quality data and iterative refinement. After optimizing the algorithm, the system screened more than 50 million molecular compounds, narrowing the field to approximately 130 high-potential repellent candidates. Laboratory and field trials validated the model accuracy. Honey bees consistently avoided the selected compounds, and subsequent field tests with freely foraging colonies confirmed that seven of the tested substances reliably repelled bees from honeycombs without causing harm. The research demonstrates a practical application of artificial intelligence in environmental conservation, offering a mechanism to shield pollinators from agricultural chemicals while maintaining crop protection. Beyond pesticide mitigation, the identified compounds present broader utility. They could reduce human-bee conflicts in public spaces such as hospitals and residential areas, and assist agricultural operations that require controlled pollination, particularly for seedless fruit varieties. The development also paves the way for next-generation, bee-friendly pesticide formulations that balance modern agricultural demands with ecosystem preservation. The research team includes Joel Kowalewski, Barbara Baer-Imhoof, Tom Guda, Matthew Luy, and Payton DePalma. Ray serves as founder and president of Sensorygen and Remote Epigenetics, with Kowalewski holding equity in Sensorygen. The inventors have filed a patent application covering the disclosed compounds. This breakthrough underscores the growing intersection of computational chemistry, behavioral biology, and sustainable agriculture, positioning machine learning as a vital tool for addressing complex ecological and agricultural challenges.