Compositionally aware estimation of cross-correlations for microbiome data.

Area of Science:

• Microbiome research
• Bioinformatics
• Statistical modeling

Background:

• Microbiome studies often analyze correlations between microbial abundances (operational taxonomic units, OTUs).
• Existing methods struggle to correlate OTU abundances with external variables or other compositional datasets.
• Compositional nature of sequencing data requires specialized analytical approaches.

Purpose of the Study:

• Introduce novel methods, SparCEV and SparXCC, for quantifying correlations between OTU abundances and external variables.
• Enable correlation analysis between OTU abundances and continuous phenotypic data or compositional datasets like transcriptomics.
• Evaluate the performance of new methods against existing transformations and cross-correlation techniques.

Main Methods:

• Developed SparCEV (Sparse Correlations with External Variables) and SparXCC (Sparse Cross-Correlations between Compositional data).
• Implemented iterative versions of SparCEV and SparXCC to address potential bias.
• Compared SparCEV/SparXCC with naive (log, log-TSS) and advanced (CLR, VST) transformations using empirical Pearson cross-correlations.

Main Results:

• CLR and VST transformations outperformed naive methods, except for dense correlation matrices.
• SparCEV and SparXCC showed superior performance over CLR/VST for small numbers of OTUs.
• Iterative procedures enhanced SparCEV/SparXCC accuracy, except in cases of near-zero average correlation or dense matrices.

Conclusions:

• SparCEV and SparXCC offer robust methods for correlating microbial abundances with external data.
• The iterative approach refines accuracy, making these methods valuable for microbiome and multi-omics studies.
• These methods advance the analysis of complex relationships within and beyond microbial communities.