Cardiovascular Disease Risk Screening for Endocrine-Disrupting Chemicals Based on Machine Learning.

Abstract

Endocrine-disrupting chemicals (EDCs) interfere with human and animal endocrine systems and are common in plastics, personal care products, and pesticides. EDC exposure has been associated with cardiovascular disease (CVD), particularly in middle-aged and older adults. We developed a computational framework to screen 1,106 EDCs for potential CVD risk, using population biomarker data on 39 environmental contaminants (ECs) from a published epidemiological study for model calibration and benchmarking. Structural clustering identified 677 candidate EDCs. A three-stage framework combining positive-unlabeled (PU) learning and stacking then identified 167 EDCs with potential CVD risk. Shapley Additive Explanations (SHAP) identified descriptors contributing to these predictions. Adverse outcome pathway (AOP) analysis, molecular dynamics simulations, and statistical methods were combined to calculate atherosclerosis effect scores and construct a priority control list. The list included 50 high-risk EDCs, with pesticides, aromatic amines, and nitrogen-containing heterocycles as prominent chemical groups. TabNet and sensitivity analysis identified molecular features associated with the predicted risk levels. The framework integrates published epidemiological data with theoretical simulations to support EDC prioritization and environmental risk management. Independent external validation is needed before applying the model to uncharacterized EDCs.

Abstract as supplied by the indexing source. © the authors / publisher.