SCNIC

SCNIC constructs correlation networks and summarizes modules of correlated features in compositional microbiome and metabolomics datasets to increase statistical power and reveal relationships among microbial communities.

Key Features:

• Correlation Network Generation: Generates correlation networks from compositional data to represent relationships among features.
• Module Detection and Summarization: Identifies modules of highly correlated features and summarizes them to reduce dimensionality.
• Louvain Modularity Maximization (LMM): Uses LMM to detect community structures within correlation networks.
• Shared Minimum Distance (SMD) Algorithm: Implements a novel SMD algorithm related to LMM and demonstrated using simulated data.
• Increased Statistical Power: Summarization of correlated features into modules enhances statistical power for downstream analyses.
• Biological Insight: Highlights microbes that differ between groups and those with strong correlations suggestive of shared environmental drivers or cooperative interactions.
• Handles Compositionality and Sparsity: Operates with consideration of microbiome data properties such as compositionality and sparsity.
• Multi-omic Versatility: Applicable to other compositional data types, including metabolomics, to support multi-omic integration.

Scientific Applications:

• Microbial interaction analysis: Identify functional relationships and co-occurrence patterns among microbes.
• Differential feature detection: Increase power to detect microbes or features that differ between experimental groups.
• Environmental influence studies: Explore shared environmental drivers of microbial community structure.
• Cooperative behavior inference: Investigate potential cooperative interactions among microorganisms via correlated modules.
• Multi-omic integration: Integrate microbiome and metabolomics compositional datasets to uncover complex biological networks.

Methodology:

Generates correlation networks from compositional data, detects modules using Louvain Modularity Maximization (LMM) and the Shared Minimum Distance (SMD) algorithm, summarizes modules as a form of dimensionality reduction to increase statistical power, and accounts for compositionality and sparsity; SMD was demonstrated using simulated data.