Unfolding and de-confounding: biologically meaningful causal inference from longitudinal multi-omic networks using...

Area of Science:

• Microbiome research
• Systems biology
• Causal inference

Background:

• Integrating multi-omic data (metagenomics, metabolomics, metatranscriptomics) is crucial for understanding microbiome interactions.
• Standard causal inference methods may overlook complex relationships and confounders in longitudinal microbiome studies.

Purpose of the Study:

• To develop and validate METALICA, a novel suite of tools for inferring biologically meaningful interactions within microbiome multi-omic datasets.
• To introduce and apply unrolling and de-confounding techniques to address limitations in existing causal inference approaches for microbiome data.

Main Methods:

• Development of METALICA, a computational tool incorporating unrolling and de-confounding algorithms.
• Application of METALICA to a longitudinal Inflammatory Bowel Disease (IBD) microbiome dataset comprising metagenomics, metabolomics, and metatranscriptomics.
• Manual validation of inferred interactions and confounders using existing literature and databases.

Main Results:

• METALICA successfully inferred interactions between microbiome entities by integrating multi-omic data.
• The unrolling process identified putative intermediary genes and metabolites explaining microbe-microbe interactions.
• The de-confounding process revealed potential common causes, clarifying spurious relationships in the IBD microbiome data.

Conclusions:

• METALICA provides a robust framework for uncovering complex multi-omic interactions in microbiome studies.
• The novel unrolling and de-confounding techniques enhance causal inference accuracy, leading to biologically validated insights.
• This approach advances our understanding of microbial communities, particularly in the context of diseases like IBD.