Table 2 Summary of best methods depending on the research question
From: A systematic benchmark of integrative strategies for microbiome-metabolome data
Scientific question | Research aim | Recommended method | Pros | Cons | Implementation |
|---|---|---|---|---|---|
Is there any relationship between microorganisms and metabolites at a global level? | Global associations | Mantel test | Robust to data normalization and distance kernels, computationally efficient | Limited ability to adjust for covariates (possible through partial Mantel tests) | vegan R pkg. |
Are microbiome and metabolome datasets summarizable through a limited number of components? | Data summarization | RDA | Robust to data normalization and distance kernels | Can only capture linear effects sensitive to directionality | vegan R pkg. |
Can we identify associations between metabolites and species? | Individual associations | MiRKAT | Allow adjustment for covariates, Robust to microbiome normalization | Limited to a few families of generalized linear models | MiRKAT R pkg. |
Can we identify core microorganisms and metabolites? | Feature selection (univariate) | CODA-LASSO (compositional covariates) | Compositional and sub-compositional coherent. No need for data transformation. Allow adjustment for covariates | Limited to a few families of generalized linear models, long running times in high-dimensional problems | coda4microbiome R pkg. |
