ViLoN

ViLoN integrates multiple molecular profiles using a multi-layer network approach to stratify patients by constructing pathway-level patient networks from gene expression, DNA methylation, and copy number variation data while incorporating prior functional knowledge from KEGG and GO.

Key Features:

• Multi-layer network approach: Integrates diverse molecular data sources via a network-based, multi-layer representation.
• Variation of Information fusion: Fuses network layers using variation of information (as reflected in the ViLoN name).
• Incorporation of prior functional knowledge: Uses KEGG and GO to inform pathway definitions and network construction.
• Patient network representation: Represents each patient as a network of pathways composed of genes linked by shared functions and joint regulation.
• Integration of data types: Integrates and is validated on combinations of gene expression, DNA methylation, and copy number variation data.
• Considers patient similarities: Accounts for similarities across patients when constructing and comparing patient networks.
• Performance in small cohorts: Demonstrates robustness in smaller cohorts where other methods may perform poorly.
• Validated stratification performance: Shows improved and consistently competitive patient stratification across different datasets.
• Clinical and research relevance: Targets patient stratification applications relevant to personalized medicine and disease-mechanism research.

Scientific Applications:

• Oncology cohort stratification: Applied to rectum adenocarcinoma and esophageal carcinoma cohorts for patient stratification.
• Identification of molecular subgroups: Identifies patient subgroups with distinct molecular characteristics to inform targeted therapeutic strategies.
• Small-sample studies: Enables stratification in smaller sample sizes where traditional methods may falter.

Methodology:

Constructs multi-layer networks fused via variation of information; integrates gene expression, DNA methylation, and copy number variation; represents patients as pathway-level networks of genes linked by shared function and joint regulation; incorporates KEGG and GO; and validates stratification across combinations of these data types while considering patient similarities.