Seurat Pipeline automated cluster detection renames values in a confusing way

[gerbeldo]

Background

For some Seurat objects, some categorical variables are detected as clustering variables, prefixing the name with the label "Cluster". For example, a Seurat object with the metadata variables "sample_name" and "patient" (which match the "group" variable heuristic, as in 1) all cells in a sample belong to at most one group of the patient variable, and 2) the cardinality of patient is less than or equal to the number of samples) returns this:

While it should return the same but without the "cluster" label.

Hypothesis

It could be related to add_samples_col function, and how it looks for a "sample" or "samples" variable and assumes a single sample if not, breaking the cardinality heuristic, and converting these "group" variables to "cluster" variables.

goal

• [ ] fix group/cluster variable detection for these cases.

[alexvpickering]

Yes, this is because there was no "sample" or "samples" column. You can confirm by checking that these groupings cannot be used for between sample DE. Some suggestions for improvements from simple to hard:

• Have some more flexible detection of "samples" metadata when the above two columns are absent (e.g. grepl('patient|sample|donor|individual', colnames(scdata@metadata)).
• Have a hard requirement on there being a metadata column that is used for sample identity (pipeline fails if fixed string or pattern not found). This would prevent the above confusing experience where the sample metadata is treated as clusters.
• Prompt the user to select a metadata column to be used for sample identity if it is not detected then re-run the pipeline.