Clarification about imputation with multiple groups

[matteobolner]

Hello everyone,
I am a PhD student currently working with metabolomics data, and I have the following situation:

I have a dataset consisting of groups and subgroups (i.e animal breed, animal sex etc).
What I need to impute is the metabolomics data, i.e numeric values, for several hundreds of metabolites with potentially missing data. Sample "metadata" is full and doesn't require any imputation.

I would like to take into account the structure of the sample groups to avoid invalid imputations.

• Should I consider the whole dataset at once, and if so, how can i meaningfully include the group information? I have tried to read the (very useful) book and resources for MICE, but it still isn't clear to me. Should I just set the sample data columns as predictors? Will the imputation take into account the group structure? It seems to be a multilevel situation, but I'm not really sure.

• Alternatively, would it be a valid procedure to split the dataset by group, impute each group and then merge the data back?

Finally, a last question:

• I have hundreds if not thousands of metabolites; it seems unfeasible to consider their properties and behavior such as distribution, convergence, etc one by one, as I often see in the examples. What would you suggest to do in my case? Do you have any references that could be of help?

I hope to have described my situation well enough to understand; please let me know if something is not clear.

Thank you again for this amazing package and all of the resources.

Matteo Bolner