Abstract

Pollinators are essential for maintaining ecosystem stability and agricultural productivity, yet their populations are in decline due to various stressors, including parasites, pathogens, climate change and pesticide exposure. Protecting pollinators has become a priority for environmental safety and food security. Regulatory authorities, including the European Food Safety Authority (EFSA), the Environmental Protection Agency (EPA) and the Organisation for Economic Co-operation and Development (OECD), have guidelines for pesticide risk assessment, but conventional testing methods are costly and time-consuming, limiting their applicability to large chemical datasets. Computational approaches, such as Structure–Activity Relationship (SAR) models, offer efficient alternatives by enabling the rapid screening of pesticides for potential toxicity to pollinators. In this study, we used a dataset of 357 compounds to develop a classification model based on structural alerts to predict oral acute toxicity in Apis mellifera. The model showed a higher Matthews Correlation Coefficient in the training set (0.82), with a moderate decay in the test set (0.56) likely due to applicability domain limits. Despite this, high balanced accuracy (0.80) and sensitivity (0.79) in the test set confirm the model as a reliable tool for the toxicological screening of pesticides.

IPC Classification

Keywords