Related Experiment Video
Updated: Jul 4, 2025
14:20
High Throughput SiRNA Screening for Chloropicrin and Hydrogen Fluoride-Induced Cornea Epithelial Cell Injury
Published on: June 16, 2018
5.9K
In silico prediction of ocular toxicity of compounds using explainable machine learning and deep learning approaches
Yiqing Zhou1, Ze Wang1, Zejun Huang1
1Shanghai Frontiers Science Center of Optogenetic Techniques for Cell Metabolism, Shanghai Key Laboratory of New Drug Design, School of Pharmacy, East China University of Science and Technology, Shanghai, China.
Journal of Applied Toxicology : JAT
|February 8, 2024
Summary
Developing computational models for chemical ocular toxicity is vital for reducing animal testing. This study created predictive models using a large dataset, with a machine learning model (RF-Descriptor) showing strong performance for practical applications.
Area of Science:
- Computational toxicology
- Chemical safety assessment
- In silico drug discovery
Background:
- Accurate identification of ocular toxicity is critical for health hazard assessment.
- There is a growing need to reduce, refine, and replace animal testing in chemical safety evaluations.
- Robust computational tools are essential for regulatory applications in toxicology.
Purpose of the Study:
- To develop and validate binary classification models for predicting chemical ocular toxicity.
- To create the most extensive dataset of chemical ocular toxicity to date.
- To balance model performance with interpretability for practical use.
Main Methods:
- Amalgamated a dataset of 4901 compounds from GHS-compliant databases and literature.
- Employed 12 molecular representations with 6 machine learning and 2 deep learning algorithms.
- Utilized SHAP and attention weights analysis for model interpretability.
Main Results:
- Deep learning model GCN achieved an AUC of 0.915 in cross-validation.
- The Random Forest model with descriptors (RF-Descriptor) was selected as the best model with an AUC of 0.869 on the test set.
- SHAP and attention weights analysis provided visual insights into key predictive features.
Conclusions:
- The developed RF-Descriptor model offers a valuable tool for predicting ocular toxicity in early drug discovery.
- The study successfully balanced predictive accuracy with model interpretability.
- The findings support the use of computational toxicology to reduce animal testing in safety evaluations.
