Classification of thyroid hormone receptor agonists and antagonists using statistical learning approaches

Area of Science:

• Computational chemistry
• Pharmacology
• Bioinformatics

Background:

• Thyroid hormone receptors (TRs) play crucial roles in metabolism and development.
• Accurate prediction of TR agonists and antagonists is vital for drug discovery and toxicological assessment.
• Existing methods for classifying TR-binding ligands have limitations.

Purpose of the Study:

• To develop and validate in silico models for predicting thyroid hormone receptor (TR) agonists and antagonists.
• To compare the performance of C4.5, Random Forest (RF), and Support Vector Machine (SVM) algorithms for this classification task.
• To establish reliable computational tools for identifying TR-binding ligands.

Main Methods:

• Utilized a dataset of 258 compounds for model development.
• Employed C4.5, Random Forest (RF), and Support Vector Machine (SVM) statistical methods.
• Validated model performance using fivefold cross-validation and an independent external test set.

Main Results:

• The C4.5 model demonstrated moderate performance with an average accuracy of 83.1% on the test set.
• The RF model achieved an average prediction accuracy of 84.0% (cross-validation) and 87.1% (external validation).
• The SVM model exhibited superior performance with overall and external prediction accuracies of 96.6% and 97.2%, respectively.

Conclusions:

• The developed in silico models, particularly RF and SVM, are reliable for classifying TR-binding ligands.
• These computational models serve as valuable tools for predicting TR agonists and antagonists.
• The findings support the utility of machine learning in drug discovery and chemical safety assessment.