CCLasso

CCLasso infers correlation networks from compositional metagenomic data to recover associations among latent variables in microbial communities.

Key Features:

• Least squares with ℓ1 penalty: Uses least squares regression with an ℓ1 penalty tailored for compositional metagenomic data to account for constraints of relative abundance.
• Compositional data handling: Operates on compositional (relative abundance) data and focuses on latent variables to avoid spurious correlations from direct Pearson analysis of relative abundances.
• Algorithmic approach: Solves the optimization problem using an alternating direction algorithm derived from the augmented Lagrangian method.
• Performance: Simulation studies report improved edge recovery relative to SparCC and comparable network estimation performance on Human Microbiome Project datasets.

Scientific Applications:

• Microbial community network analysis: Infers associations among taxa or latent variables from compositional metagenomic surveys.
• Environmental microbiomes: Characterizes interaction patterns within environmental microbial communities from compositional sequencing data.
• Human microbiome studies: Applies to human-associated datasets such as the Human Microbiome Project to recover community interaction structures.
• Disease association and therapeutics: Supports analysis of microbial interaction patterns relevant to disease association research and development of microbiome-targeted interventions.

Methodology:

Inference is based on least squares regression with an ℓ1 penalty applied to compositional data and solved via an alternating direction algorithm derived from the augmented Lagrangian method to recover correlation networks among latent variables.