Machine Learning Prediction of Transthyretin Binding for Thyroid Hormone Transport Disruption for Chemical Risk...

Area of Science:

• Toxicology and Environmental Health
• Computational Chemistry
• Endocrinology

Background:

• Endocrine-Disrupting Chemicals (EDCs) interfere with thyroid hormone (TH) homeostasis.
• Transthyretin (TTR) is crucial for TH transport, and its disruption by chemicals can impact hormone bioavailability.
• Identifying thyroidal EDCs is challenging due to co-exposures and the need for effective risk assessment tools.

Purpose of the Study:

• To develop and validate machine learning (ML)-based quantitative structure-activity relationship (QSAR) models for predicting chemical binding affinity to TTR.
• To identify molecular features associated with TTR disruption and stabilization.
• To support hazard identification and prioritization of chemicals within regulatory risk assessment frameworks.

Main Methods:

• A dataset of 1512 chemicals was used to train, test, and validate ML models predicting TTR-binding affinity.
• Feature selection involved removing correlated descriptors and ranking using mutual information regression.
• Five ML algorithms were employed, with the Gradient Boosting Regressor (GBR) model showing the best performance.

Main Results:

• The GBR model achieved high predictive accuracy (R²=0.89 training, 0.58 test, 0.55 validation).
• Analysis revealed hydrophobicity, steric effects, branching, connectivity, and ionization as key factors in TTR disruption.
• The models' applicability domain (AD) confirmed high reliability for the test and validation sets (97.5% and 96.0%, respectively).

Conclusions:

• In silico QSAR models effectively predict TTR-binding affinity, aiding in the screening of potential EDCs.
• These models provide mechanistic insights into TTR disruption, supporting non-animal testing strategies.
• The developed approach facilitates chemical prioritization for further toxicological evaluation and risk assessment.