Machine Learning-Based Toxicological Modeling for Screening Environmental Obesogens.

Area of Science:

• Environmental toxicology
• Computational chemistry
• Endocrinology

Background:

• Environmental obesogens disrupt energy balance, contributing to obesity, a significant public health concern.
• Molecular initiating events (MIEs) are crucial for understanding chemical toxicity and developing predictive models.
• Machine learning (ML) models integrating MIEs offer enhanced prediction of toxic endpoints and improved model interpretability.

Purpose of the Study:

• To construct and validate an ML-based screening system for predicting chemical obesogenic potential using MIEs.
• To identify critical molecular descriptors associated with adipogenesis and obesity.
• To assess the system's performance in predicting the obesogenic effects of substances of very high concern (SVHCs).

Main Methods:

• Integration of six MIEs related to adipogenesis and obesity into a predictive system.
• Utilizing molecular descriptors such as hydrophobicity (SlogP_VSA) and electrostatic interactions (PEOE_VSA).
• External validation using receiver operating characteristic (ROC) curves and experimental verification with 3T3-L1 adipogenesis assays.

Main Results:

• The developed system achieved high accuracy in external validation, with an area under the ROC curve of 0.78.
• Molecular hydrophobicity and direct electrostatic interactions were identified as key predictors of obesogenic potential.
• The system correctly predicted the obesogenic effects of 10 out of 12 tested SVHCs and identified four novel potential obesogens.

Conclusions:

• The integrated MIE-based screening system demonstrates robust performance in predicting adipogenic potential.
• The system effectively identifies environmental obesogens, including novel compounds, aiding in risk assessment.
• This approach enhances the understanding of chemical-induced obesity and supports regulatory efforts for substances of very high concern.