Machine learning prediction model combined with network toxicology analysis identifies potential cardiotoxic components and mechanisms among 741 pesticides

Qi Yang, Hongsheng Liu, Guiyuan Pang, Yuemi Mo, Chenglin Zhou, Meiyu Huang, Chun Luo, Erwei Hao, Lili Fan,
Machine learning prediction model combined with network toxicology analysis identifies potential cardiotoxic components and mechanisms among 741 pesticides,
Environment International,
Volume 204,
2025,
109860,
ISSN 0160-4120,
https://doi.org/10.1016/j.envint.2025.109860.
(https://www.sciencedirect.com/science/article/pii/S0160412025006117)
Abstract: The usage of pesticides in agriculture is substantial, but their application may pose potential risks of physical harm. Existing network toxicology methods for analysis and prediction lack the capability for structural early-warning of toxicity and are inefficient for individual compound analysis. The purpose of this study was to integrate a machine learning-based cardiotoxicity prediction model with network toxicology. We conducted the first cardiotoxicity prediction analysis on 741 common pesticides to improve efficiency, followed by experimental validation using a zebrafish model to confirm the model’s reliability, identify cardiotoxicity warning structures, and explore potential molecular mechanisms. We developed an optimal Random Forest (RF) cardiotoxicity prediction model achieving an AUC of 0.81; the AUC for three external cardiotoxicity validation sets all exceeded 0.70. The model predicted that 6 out of the 741 pesticides, including Copper(II) sulfate, had a cardiotoxicity probability above 0.90. Zebrafish cardiotoxicity experiments validated 6 high-probability pesticides and 5 low-probability pesticides from the predictions. The cardiomyocyte apoptosis assay achieved an 81% accuracy rate. The erythrocyte fluorescence test showed 91% accuracy, and the heart rate detection achieved 100% accuracy. Most pesticides predicted to have high cardiotoxicity potential exhibited phenomena in zebrafish such as accelerated myocardial apoptosis, reduced erythrocyte distribution in the heart, and decreased heart rate. Network toxicology analysis revealed that these pesticides likely exert their effects by targeting proteins like KCNH2 and SOD1/2, and modulating signaling pathways such as Chemical carcinogenesis and Lipid and atherosclerosis. Molecular docking results indicated strong binding affinity between the pesticides and cardiotoxicity-related targets. Toxicity structure matching showed that pesticides with high predicted cardiotoxicity possess molecular structures similar to those in cardiotoxicity databases. The integrated machine learning and network toxicology approach developed in this study is reliable. Pesticides predicted to have potential cardiotoxicity require prioritized prevention during use.
Keywords: Pesticides; Cardiotoxicity; Machine learning; Network toxicology; Zebrafish