CCLasso: correlation inference for compositional data through Lasso.

Area of Science:

• Microbiology
• Bioinformatics
• Computational Biology

Background:

• High-throughput sequencing of 16S rRNA genes enables microbial community analysis.
• Inferring microbial correlations is crucial for genomic studies.
• Compositional data from sequencing requires specialized correlation methods.

Purpose of the Study:

• To propose a novel method for inferring correlation networks from compositional metagenomic data.
• To address limitations of traditional correlation analyses on relative abundance data.

Main Methods:

• Developed CCLasso, a method using least squares with L1 penalty for compositional data.
• Employed an alternating direction algorithm from augmented Lagrangian for optimization.
• Validated CCLasso using simulation studies and Human Microbiome Project data.

Main Results:

• CCLasso demonstrates superior edge recovery compared to existing methods like SparCC.
• CCLasso effectively estimates microbe species correlation networks.
• Performance is comparable to SparCC on real-world metagenomic data.

Conclusions:

• CCLasso provides a robust approach for analyzing microbial community structures.
• The method enhances the reliability of correlation inference in metagenomics.
• Open-source availability facilitates broader application in microbial ecology research.