Deep Learning-Driven Discovery of Bee-Safe Isoxazoline Pesticide Candidates

Isoxazoline pesticides, such as fluxametamide, while effective against parasites and pests, pose a severe environmental threat due to their high toxicity to honeybees - critical pollinators essential for ecosystem health and food security. Existing predictive platforms fail to accurately assess this risk for isoxazolines due to critical data gaps. To address this issue, we developed <b>BeeSafe 2.0</b>, an innovative deep learning model uniquely integrating graph neural networks (GGHT) and residual networks (ResNet) architecture, further enhanced by new training set. <b>BeeSafe 2.0</b> demonstrates superior predictive performance, specifically overcoming previous limitations for isoxazolines, and provides an accessible online server (www.beesafe.top) for chemical bee toxicity assessment. Crucially, leveraging <b>BeeSafe 2.0</b>, we discovered <b>WT-02</b>, a novel isoxazoline insecticide exhibiting potent efficacy against diverse pests while displaying dramatically reduced bee toxicity (only 1/18 of that for fluxametamide). This work presents a transformative "new architecture-new data-application" approach, offering a powerful tool for environmental risk management of pesticides and enabling the discovery of truly bee-safe, greener alternatives to safeguard pollinator health and promote sustainable agriculture.