A Machine Learning Approach to Molecular Initiating Event Prediction Using High-Throughput Transcriptomic Chemical...

Area of Science:

• Toxicogenomics
• Computational Biology
• Bioinformatics

Background:

• High-throughput RNA sequencing offers potential for chemical hazard identification but faces challenges due to data complexity.
• Extracting mechanistic insights from large transcriptomic datasets is crucial for chemical safety assessment.

Purpose of the Study:

• To develop and validate a machine learning framework (MIEML) for predicting molecular initiating events (MIEs) from transcriptomic chemical bioactivity screens.
• To address the need for bioinformatic approaches to utilize high-dimensional transcriptomic data for chemical safety.

Main Methods:

• Trained machine learning classifiers to predict MIE activation using gene expression profiles from MCF-7 cells exposed to chemicals.
• Integrated transcriptomic data with chemical-MIE annotations from the RefChemDB database.
• Validated classifier performance against known MIEs and applied predictions to a large set of test chemicals.

Main Results:

• Three distinct MIE classifiers demonstrated statistically significant predictive performance (p ≤ 0.1) compared to random models.
• The validated classifiers accurately predicted MIEs for chemicals not included in the training set.
• Predictions for 1750 test chemicals showed substantial agreement with results from targeted molecular bioassays for key receptor activities.

Conclusions:

• The MIEML framework effectively predicts molecular initiating events from large-scale transcriptomic data.
• This approach enhances the utility of RNA sequencing for chemical hazard identification and safety assessment.
• Demonstrates a viable computational strategy for leveraging transcriptomic data to fill gaps in chemical safety information.