Drug candidate identification based on gene expression of treated cells using tensor decomposition-based unsupervised...

Area of Science:

• Computational Biology
• Bioinformatics
• Drug Discovery

Background:

• Traditional in silico drug discovery methods (structure-based, ligand-based) have limitations.
• Current gene expression-based approaches require training datasets for drug-protein interaction inference.
• A novel approach is needed to overcome these limitations in drug development.

Purpose of the Study:

• To develop a new in silico drug discovery approach.
• To infer drug-protein interactions without pre-existing knowledge.
• To integrate gene expression profiles with gene-gene interaction data.

Main Methods:

• Utilized tensor decomposition-based unsupervised feature extraction.
• Screened gene sets and compounds for dose-dependent activity without training datasets.
• Integrated results with single-gene expression perturbation data to infer target genes.

Main Results:

• Successfully screened compounds and gene sets using a novel unsupervised method.
• Inferred target genes for analyzed compounds with significant overlap to known targets.
• Demonstrated the method's effectiveness on large-scale datasets.

Conclusions:

• Identified WZ-3105 and CGP-60474 as promising drug candidates.
• These compounds show potential for targeting multiple cancers including melanoma, adenocarcinoma, liver, breast, colon, and prostate cancers.
• The developed method is suitable for large-scale drug discovery datasets.