antagomir
commented
3 months ago Thanks!
Regression is indeed a common statistical technique.
The primary focus of OMA is to teach Bioconductor methods that support the modern multi-assay data containers, in particular the (Tree)SummarizedExperiment and MultiAssayExperiment but possibly others. OMA is not a book about general statistics (a topic which has more comprehensive treatments elsewhere). A key shortcoming in this example is that it does not show how to do regression on such data objects.
Another gap is in the statistical assumptions; read counts or relative abundances in microbiome context usually violate assumptions of standard linear regresssion in multiple ways and that is pedagogically not ideal. Examples on GLMs would be better justified but for those we do have DA tests already available for individual taxa.
If we keep linear regression example then I would implement following changes:
- use one of the microbiome demo data sets
- regress on alpha diversity (Shannon index?) instead of individual taxa (as we already have more dedicated DA methods available for taxa)
- emphasize linear modeling assumptions and include some testing on how well the assumptions hold
- show how to treat possible covariates
- emphasize visualization and interpretation of the analysis outcomes
- where possible, use visualization methods that are dedicated to (Tree)SE/SCE objects, e.g. from miaViz, scater or elsewhere
jkc9886
One of the methods to visualize microbiome data in the OMA book chapters is regression charts, however there was no example or mention about it in the exercises chapter.
I have added:
Please suggest if this could be added and if yes, what better could be done?