Drug vector representation: a tool for drug similarity analysis

Area of Science:

• Toxicogenomics
• Bioinformatics
• Computational Biology

Background:

• DrugMatrix dataset offers in vivo transcriptome data for numerous drugs.
• Understanding drug relationships and biological effects remains challenging.
• High-dimensional microarray data limits direct application and analysis.

Purpose of the Study:

• To represent transcriptome data using lower-dimensional vectors.
• To enable drug similarity comparisons and infer drug mechanisms of action (MoA).
• To model drug combinations and predict genotoxicity features.

Main Methods:

• Latent Semantic Analysis (LSA) from natural language processing was adapted.
• Treatments were represented by dense, orthogonal biological feature vectors.
• Cosine similarity calculated for treatment vectors; gProfiler for annotation.

Main Results:

• Vector representations facilitated drug similarity comparisons.
• Drug-drug interaction pairs showed higher vector similarity.
• Vector features aided in elucidating drug MoA and genotoxicity.
• A web tool was developed for exploring treatment similarities and combinations.

Conclusions:

• Vector-based representation of transcriptome data simplifies complex toxicogenomic datasets.
• The developed tool supports hypothesis generation for drug repurposing and combination therapy.
• This approach enhances the utility of the DrugMatrix dataset for toxicogenomic research.