RWRF

RWRF integrates multi-omics datasets by applying a Random Walk with Restart (RWR) algorithm on a multiplex sample network to fuse similarity networks and enable identification of molecular subtypes.

Key Features:

• Multiplex Network Construction: Constructs individual similarity networks for each omics data type, representing relationships between samples for mRNA expression, DNA methylation, and microRNA expression.
• Integration via Multiplex Sample Networks: Connects corresponding samples across individual similarity networks to form a comprehensive multiplex sample network that represents multiple data dimensions simultaneously.
• Random Walk with Restart Algorithm: Applies the Random Walk with Restart (RWR) algorithm on the multiplex network to propagate similarity information across layers.
• Stationary Probability Distribution Utilization: Uses the stationary probability distribution from the RWR process as the basis for fusing integrated similarity networks.
• Network Clustering: Performs network clustering on the fused similarity network to identify molecular subtypes within cancer datasets.
• Performance Evaluation: Demonstrates, on TCGA cancer datasets, improved identification of cancer subtypes compared to single-data analyses and prior integrative methods.

Scientific Applications:

• Precision Medicine: Integrates multiple omics layers to support analyses relevant to precision medicine by revealing cross-omics molecular interactions.
• Cancer Subtype Discovery: Applied to The Cancer Genome Atlas (TCGA) datasets integrating mRNA expression, DNA methylation, and microRNA expression to identify cancer molecular subtypes.

Methodology:

Construct individual similarity networks per omics type; connect corresponding samples across networks to form a multiplex sample network; apply Random Walk with Restart (RWR) on the multiplex network and use the stationary probability distribution to fuse integrated similarity networks; perform network clustering to identify molecular subtypes; evaluate performance against single-data and prior integrative methods on TCGA datasets.