Discovering functional modules by topic modeling RNA-Seq based toxicogenomic data.

Area of Science:

• Computational biology
• Toxicogenomics
• Bioinformatics

Background:

• Toxicogenomics (TGx) aims to understand molecular mechanisms of toxicity.
• RNA-Seq is a powerful tool for TGx, but data analysis presents challenges.
• Novel methods are needed to extract insights from complex RNA-Seq TGx data.

Purpose of the Study:

• To apply topic modeling to RNA-Seq based TGx data for discovering hidden functional modules.
• To explore the utility of topic modeling in clustering samples by compound modes of action (MoAs).
• To validate the transferability of identified topics from RNA-Seq to microarray data.

Main Methods:

• RNA-Seq gene expression profiles were treated as "documents" for text mining.
• Latent Dirichlet Allocation (LDA) was used to build a topic model.
• Gene enrichment analysis and literature mining were used to interpret "marker" topics.

Main Results:

• Samples treated with compounds having similar MoAs clustered together based on topic similarity.
• Identified "marker" topics were successfully interpreted and validated.
• Topic transferability from RNA-Seq to microarrays achieved approximately 85% accuracy for a specific pathway.

Conclusions:

• Topic modeling is applicable for discovering functional modules in RNA-Seq TGx data.
• This approach provides a valuable computational tool for leveraging TGx data in the RNA-Seq era.
• The method demonstrates potential for cross-platform data analysis and MoA identification.