Machine learning of honey bee olfactory behavior identifies repellent odorants in free-flying bees in the field.

Preventing beneficial insects like honey bees (Apis mellifera) from contacting pesticides on crops using odorants could counter current pollinator declines. However, the discovery of behaviorally aversive odorants is impeded by the complexity of the honey bee olfactory system where >170 olfactory receptors detect volatiles and generate valence. To solve this systems-level challenge, we generated a machine-learning model to predict aversive valence from chemical structure using published olfactory behavior data in honey bees. We refine the predictive model by generating species-level behavioral data for honey bees and Drosophila on an initial set of novel predicted repellents. The improved second computational model was then used to screen a chemical space of >50 million compounds and identify >130 repellent candidates. Behavioral validation using honey bees in the laboratory shows a high predictive success. Additional testing of the top seven candidates using freely foraging honey bees in a field assay confirmed strong repellency, thus predicting a high probability to repel foraging bees from pesticide-treated crops. Machine learning, with iterative testing and modeling, therefore provides a powerful approach for rational discovery of aversive volatiles for control of insects for which limited data is available.