An elastic network model to identify characteristic stress response genes

Area of Science:

• Toxicology
• Systems Biology
• Bioinformatics

Background:

• Gene expression profiling can predict toxic effects, but similar responses across different toxic compounds complicate predictions.
• A global stress response affects numerous genes, making it challenging to identify specific toxicity indicators.

Purpose of the Study:

• To develop a predictive model for toxic effects using gene expression data.
• To identify reliable biomarkers for toxicity prediction despite global gene expression changes.

Main Methods:

• An elastic network model was created, representing genes as nodes and coexpression as edges.
• The model predicts global gene expression responses by analyzing a small subset of gene expression data.
• A neural network was trained to classify compound toxicity based on profiling experiments.

Main Results:

• The elastic network model effectively captures global gene expression patterns, requiring only a few gene expressions to model the entire response.
• Identified key genes and atypical responders proved to be effective biomarkers for toxicity prediction.
• A neural network achieved 87.5% accuracy (14 out of 16 compounds correctly classified) in predicting toxicity on a validation dataset.

Conclusions:

• Elastic network models can identify biologically plausible biomarkers for toxicity prediction.
• This approach effectively distinguishes subtle biological signals within broad response backgrounds.
• The methodology is adaptable for biomarker discovery in other biological research areas beyond toxicology.