Leveraging In Silico Structure–Activity Models to Predict Acute Honey Bee (Apis mellifera) Toxicity for Agrochemicals

In the realm of crop protection products, ensuring the
safety of
pollinators stands as a pivotal aspect of advancing sustainable solutions.
Extensive research has been dedicated to this crucial topic as well
as new approach methodologies in toxicity testing. Hence, within the
agricultural and chemical industries, prioritizing pollinator safety
remains a constant objective during the development of predictive
tools. One of these tools includes computational models like quantitative
structure–activity relationships (QSARs) that are valuable
in predicting the toxicity of chemicals. This research uses bee toxicity
data to develop artificial neural network classification models for
predicting honey bee acute toxicity. Bee toxicity data from 1542 compounds
were used to develop models; the sensitivity and specificity of the
best model were 0.90 and 0.91, respectively. These in silico models
can aid in the discovery of next-generation crop protection products.
These tools can guide the screening and selection of next-generation
crop protection molecules with high margins of safety to pollinators,
and candidates with favorable sustainability profiles can be identified
at the early discovery stage as precursors to in vivo data generation.