A Qualitative Modeling Approach for Whole Genome Prediction Using High-Throughput Toxicogenomics Data and...

Area of Science:

• Toxicogenomics
• Computational Biology
• Environmental Health

Background:

• High-throughput transcriptomics (HTT) offers rapid chemical toxicity assessment but relies on models trained with pharmaceutical data, potentially limiting generalizability.
• Existing HTT methods may not be suitable for diverse chemical spaces found in commercial and environmental contaminants.
• Predicting genome-wide gene expression from limited data is crucial for efficient toxicity screening.

Purpose of the Study:

• To develop and validate predictive computational models for inferring whole genome transcriptional profiles from a small set of surrogate genes.
• To assess the suitability of models trained on diverse chemical exposures for general toxicity testing.
• To compare the predictive performance of novel surrogate gene sets against existing landmark gene sets.

Main Methods:

• Trained and validated predictive models using the Open TG-GATEs toxicogenomics database, including human primary hepatocyte data from 158 compounds.
• Employed a sequential forward search-based greedy algorithm with machine learning techniques to identify an optimal set of surrogate genes (SV2000).
• Assessed predictive performance by comparing the SV2000 set against L1000 and S1500 landmark genes and analyzing pathway enrichment.

Main Results:

• Developed predictive models capable of inferring whole genome transcriptional profiles from a limited set of surrogate genes.
• The SV2000 surrogate gene set demonstrated superior predictive performance compared to existing L1000 and S1500 landmark genes.
• Pathway enrichment analysis confirmed the ability of the limited gene set to determine relevant tissue response patterns.

Conclusions:

• The developed SV2000 gene set and predictive models enhance the accuracy and efficiency of high-throughput transcriptomics for chemical safety evaluation.
• These findings support the incorporation of mode of action (MOA) analysis into high-throughput chemical prioritization and testing.
• The approach offers a robust method for predicting transcriptional responses across different species and data platforms.