Combining compositional data sets introduces error in covariance network reconstruction.

Area of Science:

• Microbiology
• Bioinformatics
• Ecology

Background:

• Microbial communities comprise diverse taxa across multiple kingdoms.
• Interactions between bacteria and fungi significantly influence community structure.
• Inferring cross-kingdom associations is more complex than intra-kingdom associations due to data characteristics.

Purpose of the Study:

• To quantify the theoretical and practical challenges of cross-kingdom network inference.
• To evaluate common network inference techniques for their ability to handle combined compositional datasets (e.g., 16S and ITS sequencing).
• To assess the accuracy and usefulness of intra- and inter-kingdom associations derived from these techniques.

Main Methods:

• Utilized synthetic and real-world microbiome data.
• Detailed the theoretical issues of combining compositional datasets from the same environment.
• Surveyed and tested common network inference techniques using simulated samples with known ground-truth associations.
• Identified error signatures in transkingdom network inference.

Main Results:

• Standard network inference techniques show limitations in accurately inferring cross-kingdom microbial associations.
• While methods mitigate some cross-kingdom inference errors, differences are minimal for practical applications like identifying strong correlations or keystone taxa.
• A distinct error signature from transkingdom inference was identified and observed in real-world environmental microbiome data.

Conclusions:

• Cross-kingdom network inference from combined microbiome datasets presents significant challenges for current standard techniques.
• The accuracy and utility of identifying microbial interactions and keystone taxa across kingdoms remain limited.
• The identified error signature provides a means to recognize potential inaccuracies in transkingdom microbiome network analyses.