BeeMSAI: A multi-dimensional evaluation strategy for honeybee toxicity in agrochemicals integrating mode of action,...

Honeybees (Apis mellifera) play a crucial role in crop pollination; however, their population is decreasing owing to pesticide exposure. Understanding the mechanisms underlying pesticide toxicity and accurately predicting toxicity to honeybees are essential for developing sustainable agrochemicals and effective ecological risk management strategies. Herein, we developed a multi-dimensional evaluation strategy combining mode of action (MOA), structural analysis, and artificial intelligence (AI)-powered prediction (BeeMSAI) to analyze the toxicity of pesticides to honeybees. First, an MOA-based classification system was established for pesticides, which were divided into 46 MOAs and 9 physiological systems. Second, toxicity classification rules were established for six high-variability structural groups. Six representative structures with a toxicity probability of > 97.8 % were identified through fragment analysis. Third, a series of ternary classification AI models for predicting pesticide toxicity to honeybees was developed through multi-feature fusion. The support vector machine model showed robust predictive performance with an overall accuracy of 81.82 %, with the accuracy for neonicotinoids and new pesticides being 84.21 % and 80.77 %, respectively. Furthermore, an online user-friendly platform was developed to assess agrochemical risks, achieving a recall level approximately 35 % higher than that of existing online platforms. Overall, this study introduces a multi-dimensional strategy for assessing pesticide toxicity to honeybees, providing insights for developing ecologically safe agrochemicals.