Related Experiment Video
Updated: Jul 29, 2025
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
Machine Learning Model for Screening Thyroid Stimulating Hormone Receptor Agonists Based on Updated Datasets and
Wenjia Liu1, Zhongyu Wang1, Jingwen Chen1
1Key Laboratory of Industrial Ecology and Environmental Engineering (Ministry of Education), Dalian Key Laboratory on Chemicals Risk Control and Pollution Prevention Technology, School of Environmental Science and Technology, Dalian University of Technology, Dalian 116024, China.
New machine learning models effectively screen thyroid stimulating hormone receptor (TSHR) agonists, crucial endocrine-disrupting chemicals (EDCs). This study introduces an enhanced dataset and applicability domain (AD) characterization for improved chemical safety assessments.
Area of Science:
- Computational chemistry
- Toxicology
- Machine learning
Background:
- Screening endocrine-disrupting chemicals (EDCs) like thyroid stimulating hormone receptor (TSHR) agonists is vital for chemical safety.
- Previous machine learning (ML) models for TSHR agonists suffered from imbalanced datasets and lacked regulatory-applicable domain (AD) characterization.
Purpose of the Study:
- To develop improved ML models for screening TSHR agonists with enhanced datasets and robust AD characterization.
- To establish a novel AD characterization methodology (ADSAL{ρs, IA}) for ML models in chemical screening.
Main Methods:
- An updated TSHR agonist dataset was created, balancing active and inactive compounds (1:2.6 ratio) and enhancing structure-activity landscapes (SALs).
- Seven molecular representations and four ML algorithms were evaluated.
- A new methodology using weighted similarity density (ρs) and weighted inconsistency of activities (IA) was developed for SAL characterization and AD definition.
Main Results:
- Optimized ML models significantly outperformed previous ones.
- The optimal classifier, using PubChem fingerprints and random forest, achieved an AUC of 0.984 and balanced accuracy of 0.941.
- The established AD methodology (ADSAL{ρs ≥ 0.15, IA ≤ 0.65}) successfully characterized model applicability and identified 90 novel TSHR agonist classes.
Conclusions:
- The developed ML classifier and ADSAL methodology offer efficient tools for screening EDCs, specifically TSHR agonists.
- The AD characterization approach is adaptable for other ML models in chemical safety assessment.
