Identification of early liver toxicity gene biomarkers using comparative supervised machine learning
Brandi Patrice Smith1,2, Loretta Sue Auvil3, Michael Welge3,4
1Department of Food Science and Human Nutrition, University of Illinois, 1201 W Gregory Dr, Urbana-ChampaignUrbana, IL, 61801, USA.
Scientific Reports
|November 6, 2020
Summary
This study identifies ten gene biomarkers to predict liver necrosis and potential liver cancer. These biomarkers can accelerate toxicity testing for agrochemicals and pharmaceuticals, reducing costs and time.
Area of Science:
- Toxicology
- Genomics
- Biomarker Discovery
Background:
- Regulatory approval for agrochemicals and pharmaceuticals requires liver toxicity screening, a costly and lengthy process.
- Early exposure gene signatures and predictive models offer potential to reduce toxicity testing time and expenses.
Purpose of the Study:
- To develop and validate a gene signature for predicting liver toxicity using machine learning.
- To identify reliable biomarkers for early detection of liver necrosis and potential carcinogenesis.
Main Methods:
- Comparative supervised machine learning approaches were applied to the rat liver TG-GATEs dataset.
- Three different feature selection methods were used to identify gene biomarkers.
- An independent validation dataset from the Microarray Quality Control (MAQC)-II study was utilized.
Main Results:
- Ten gene biomarkers were identified that accurately predicted liver necrosis with high specificity and selectivity.
- Nine of the ten genes are involved in metabolism, detoxification, and transcriptional regulation.
- Several identified genes (Crat, Car3, Slc23a1) are implicated in liver carcinogenesis.
Conclusions:
- A biomarker gene signature with high statistical accuracy and a manageable number of genes was developed.
- This signature can potentially accelerate toxicity testing for liver necrosis and cancer.
- The findings support the use of gene signatures in regulatory toxicity testing to improve efficiency and reduce costs.
