Applying knowledge-driven mechanistic inference to toxicogenomics.

Area of Science:

• Toxicology
• Computational Biology
• Biochemistry

Background:

• Understanding chemical toxicity mechanisms is crucial but resource-intensive, often relying on animal models.
• Current predictive toxicology methods may lack detailed mechanistic insights.
• A need exists for efficient tools to generate causal explanations for toxicity.

Purpose of the Study:

• To introduce MechSpy, an innovative mechanistic inference framework for hypothesis generation in toxicology.
• To provide a tool that combines human biology, toxicology, and biochemistry knowledge with gene expression data.
• To offer a mechanistic explanation for predicted toxicities, moving beyond statistical predictions.

Main Methods:

• Developed a mechanistic inference framework (MechSpy) integrating a semantically-interconnected knowledge base.
• Utilized gene expression time series data from human tissues.
• Employed vector representations of biological entities to identify enrichment in curated toxicity mechanisms (ontology concepts).

Main Results:

• MechSpy successfully predicted canonical toxicity mechanisms for well-studied compounds.
• Experimental validation confirmed predictions for chemicals with previously unknown toxicity mechanisms.
• The framework provides mechanistic explanations for each prediction.

Conclusions:

• MechSpy serves as an advantageous tool for predictive toxicology, offering mechanistic insights.
• It is the first framework to provide mechanistic explanations for toxicity predictions.
• The framework is adaptable for additional toxicity mechanisms and generalizable to other biological processes.