Adding pseudobulk-level covariates defined across cell types, to a DE/DS analysis

[koenvandenberge]

Suppose I am working with a large study with lots of heterogeneity, and I calculate factors of unwanted variation (UV) using RUVSeq using all pseudobulk samples (several cell types, 2 conditions, several patients).

In a muscat workflow, the SCE object upon aggregation, has colData that is of dimension equal to the number of patients. However, the UV factors are of dimension equal to number of cell types x number of conditions x number of patients ( = total number of pseudobulk samples). It seems reasonable to include the factors of unwanted variation for the relevant samples in a muscat analysis, i.e., extracting the relevant elements of a factor of UV for each of the cell type-specific comparisons, and adding it to the model matrix.

The only way I can see this is possible currently is by manually implementing the edgeR analysis: aggregating counts, extracting relevant (elements of) covariates, and running the edgeR workflow. Is there a better way to accommodate analyses like these using muscat?

[koenvandenberge]

This may be related to #50.