Joint-SNF

Joint-SNF integrates multi-omics data using Joint and Individual Variation Explained (JIVE) and Similarity Network Fusion (SNF) to identify molecular subtypes of lower-grade gliomas (LGG) for improved diagnosis and therapeutic stratification.

Key Features:

• Integration of Multi-Omics Data: Integrates mRNA expression, DNA methylation, and microRNA (miRNA) data to produce a combined molecular profile of LGG samples.
• Extraction of Joint Structures (JIVE): Extracts joint variation across omics datasets using Joint and Individual Variation Explained (JIVE) to emphasize shared signals.
• Similarity Network Fusion (SNF): Applies Similarity Network Fusion to integrate joint structures into a cohesive fused similarity network.
• Enhanced Network Construction: Constructs a fused network that enhances true sample similarities while reducing noise and spurious associations.
• Spectral Clustering for Subtype Identification: Uses spectral clustering on the fused network to identify molecular subtypes among LGG patients.
• Superior Performance in Simulations: Simulation studies demonstrated that Joint-SNF outperforms the original SNF approach across various scenarios.
• Clinical Relevance: In a Chinese LGG cohort, Joint-SNF identified three molecular subtypes with five-year mortality rates of 80.8%, 32.1%, and 34.4%, and after adjustment for clinical covariates Cluster 1 showed a 5.06-fold higher risk of death compared with other clusters.

Scientific Applications:

• Molecular Subtyping of LGG: Defines molecular subtypes of lower-grade gliomas (LGG) to inform prognosis and treatment stratification.
• Prognostic Stratification: Stratifies patients by survival risk based on subtype-specific survival differences.
• Biomarker Discovery and Precision Oncology: Supports biomarker discovery and the selection of targeted therapies through integrated multi-omics subtype identification.

Methodology:

Joint-SNF extracts joint structures via JIVE, integrates those structures using Similarity Network Fusion to construct a fused similarity network, and applies spectral clustering for molecular subtype identification; performance was evaluated using simulation studies.