Multi-layered network-based pathway activity inference using directed random walks: application to predicting...

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Comprehensive analysis of cancer molecular features requires multi-omics data integration.
• Pathway activity inference methods facilitate understanding of multi-gene effects.
• Previous iDRW method showed promise but lacked generality.

Purpose of the Study:

• To develop a generalized, multi-layered network approach for pathway activity inference.
• To improve outcome prediction and biological insight from cancer multi-omics data.
• To identify common and cancer-specific driver pathways as prognostic features.

Main Methods:

• Designed a directed gene-gene graph incorporating pathway information across multiple network layers.
• Utilized a proof-of-concept study with three genomic profiles of urologic cancer patients.
• Implemented the iDRW approach as an R software package.

Main Results:

• Achieved superior outcome prediction performance compared to single genomic profiles and existing methods.
• Identified common and cancer-specific candidate driver pathways as predictive prognostic features.
• Provided enhanced biological insights through an integrated, multi-layered gene-gene network view.

Conclusions:

• The generalized integrative approach significantly improves outcome prediction in cancer.
• The method successfully identifies key driver pathways with prognostic value.
• The framework is broadly applicable to various datasets beyond urologic cancers.