Researchers at the University of California, Riverside, have used machine learning to identify odor compounds that repel honey bees, with the goal of keeping bees away from pesticide-treated crops, according to a paper in the journal eLife (opens in new tab) and a UC Riverside news (opens in new tab) release published Sept. 24, 2026.
The team, led by molecular biologist Anandasankar Ray and working with entomologist Boris Baer's lab, trained a model on the chemical structures of odorants and on bee behavioral data, then refined it with new lab data from honey bees and fruit flies. The model screened more than 50 million compounds and flagged more than 130 candidate repellents, the paper says.
In lab tests, honey bees avoided the top candidates, largely as the model predicted. In field tests with freely foraging bees, all seven compounds tested repelled bees from treated honeycomb foundations, the paper reports. UC Riverside's release describes the compounds as repelling bees "safely" and "without harming them"; those characterizations come from the university.
The researchers say the compounds could be added to pesticides to reduce bee exposure and could also be used to discourage hives in buildings or to limit pollination of seedless fruit crops. Neither the paper nor the release reports an approved product.
The authors disclosed financial interests. According to the UC Riverside release, Ray is the founder and president of Sensorygen and Remote Epigenetics and holds equity in both companies, and co-author Joel Kowalewski holds equity in Sensorygen; the eLife (opens in new tab) competing-interest statements also list co-author Matthew Luy as holding Sensorygen equity and describe Sensorygen as a startup working on insect repellents. Six of the seven authors, including Ray and Baer, are inventors on a UC Riverside patent application covering the compounds. The research was funded by a grant from the California Research Alliance by BASF; the paper states that the funder had no role in study design, data collection or the decision to publish.
The study went through eLife (opens in new tab)'s reviewed-preprint process: it was first posted as a preprint in November 2024, has been available as a reviewed preprint since January 2025, and was published as a final version of record on Sept. 23, 2026. eLife (opens in new tab)'s published assessment, written by its editors and reviewers, rated the findings "important" and the evidence "solid."