Study for the binding affinity of thyroid hormone receptors based on machine learning algorithm.

Area of Science:

• Environmental Chemistry
• Toxicology
• Computational Chemistry

Background:

• Long-term exposure to exogenous compounds can disrupt thyroid hormone receptors (TRs), potentially leading to thyroid dysfunction.
• Quantitative structure-activity relationship (QSAR) modeling offers a predictive approach to assess the binding affinity of compounds to TRs.

Purpose of the Study:

• To develop and validate robust QSAR models for predicting the binding affinity (pIC50) of compounds to thyroid hormone receptor alpha (TRα) and beta (TRβ).
• To identify key molecular descriptors influencing the interaction of compounds with TRα and TRβ.
• To apply the developed models for predicting the TR binding affinities of known endocrine-disrupting compounds.

Main Methods:

• Compilation of comprehensive datasets for TRα and TRβ binding affinities.
• Development and evaluation of five machine learning models, including support vector regression.
• Rigorous model validation using various evaluation parameters to ensure reliability and predictive power.

Main Results:

• The support vector regression model demonstrated superior robustness and external prediction ability for both TRα (r²train=0.77, r²test=0.78) and TRβ (r²train=0.78, r²test=0.80).
• Key molecular features influencing TRα binding included molecular volume, mass, and aromaticity.
• Molecular weight, electrical properties, and hydrophilicity were significant factors for TRβ binding affinity.
• The models were successfully applied to predict the TR binding affinities of 109 endocrine disruptors.

Conclusions:

• Developed QSAR models, particularly support vector regression, serve as effective tools for predicting TR binding affinity.
• These models can aid in identifying potential thyroid disruptors among exogenous compounds.
• The identified molecular descriptors provide insights into the mechanisms of TR binding, facilitating the design of safer chemicals.