NetSHy

NetSHy summarizes biological networks at the subject level by integrating node profiles with network topology using principal component analysis (PCA) and the Laplacian matrix to improve dimensionality reduction for phenotype association analyses.

Key Features:

• Hybrid PCA–Laplacian integration: Combines principal component analysis (PCA) with the graph Laplacian matrix to incorporate topological similarity into dimensionality reduction.
• Node profile and topology integration: Integrates node profiles with topological information derived from the network's structure to produce subject-level summarizations.
• Subject-level summarization: Generates subject-level network summarization scores for downstream analyses.
• Improved phenotype correlation recovery: Recovers true correlations with phenotypes more accurately than conventional PCA applied solely to node profiles.
• Higher explained variation for sparse networks: Maintains higher explained variation in data, particularly for sparse networks.
• Robustness to reduced sample size: Provides consistent correlation results as sample sizes decrease, indicating robustness to smaller cohorts.
• GWAS application: Applied in genome-wide association studies to identify more significant single nucleotide polymorphisms (SNPs) compared to traditional network representations.
• Simulation validation: Validated using simulation studies on random and empirical networks across varying sizes and sparsity levels.

Scientific Applications:

• GWAS SNP discovery: Enhances detection of single nucleotide polymorphisms in genome-wide association studies by using network summarization scores.
• Phenotype association analyses: Improves association analyses between network-derived features and phenotypic traits by better recovering true correlations.
• Dimensionality reduction for modular networks: Provides dimensionality reduction and summarization for modular biological networks with subnetworks or modules to facilitate system-level interpretation.

Methodology:

Integrates node profiles with a Laplacian matrix representation of network topology and applies principal component analysis (PCA) to derive subject-level summarization scores; performance was evaluated via simulation studies on random and empirical networks of varying size and sparsity and by application to GWAS for SNP association.