Related Experiment Video
Updated: Oct 19, 2025
A High-throughput Assay for the Prediction of Chemical Toxicity by Automated Phenotypic Profiling of Caenorhabditis elegans
Published on: March 14, 2019
Machine learning-based biomarkers identification from toxicogenomics - Bridging to regulatory relevant phenotypic
Sheikh Mokhlesur Rahman1, Jiaqi Lan2, David Kaeli3
1Department of Civil and Environmental Engineering, Northeastern University, 360 Huntington Ave, Boston, MA 02115, USA; Department of Civil Engineering, Bangladesh University of Engineering and Technology, Dhaka 1000, Bangladesh.
This study uses machine learning to identify key biomarkers for predicting toxicity. This approach reduces redundant markers, improving the accuracy and efficiency of toxicity screening for environmental and health applications.
Area of Science:
- Toxicology and Computational Biology
- Biomarker Discovery
- Predictive Toxicology
Background:
- Establishing quantitative links between in vitro molecular assays and in vivo toxicity endpoints is crucial for the Tox21 vision.
- Current toxicomics approaches often use numerous redundant markers, increasing costs and complexity.
- Selecting minimal, relevant biomarkers is essential for efficient toxicity screening and risk monitoring.
Purpose of the Study:
- To develop a method for identifying an optimal set of minimally redundant biomarkers for predicting in vivo toxicity.
- To apply machine learning techniques for feature selection and classification in toxicomics.
- To demonstrate the utility of this approach using case studies for carcinogenicity and genotoxicity prediction.
Main Methods:
- Utilized time-series toxicomics in vitro assays.
- Employed maximum relevance and minimum redundancy (MRMR) for feature selection.
- Applied support vector machine (SVM) for classification.
- Validated the approach with two case studies: in vivo carcinogenicity and Ames genotoxicity.
Main Results:
- Identified an "optimal" number of biomarkers with minimal redundancy for accurate toxicity prediction.
- Achieved 76% accuracy (AUC=0.81) for in vivo carcinogenicity prediction using five biomarkers.
- Achieved 70% accuracy (AUC=0.75) for Ames genotoxicity prediction using five biomarkers.
- Discovered distinct sets of top-ranked biomarkers for each endpoint, related to DNA repair pathways.
Conclusions:
- A small ensemble of properly selected biomarkers, particularly those involved in DNA damage and repair pathways, can accurately predict phenotypic toxicity endpoints.
- The developed method addresses the knowledge gap in phenotypic anchoring and predictive toxicology.
- This approach supports the advancement of the Tox21 vision for environmental and health applications.
