pbDS removes samples not satisfying min_cells / outliers requirements. The checks in pbDS.R:147-149 check that after these exclusions, it is still possible to fit the full model.
This means that where a cluster has many cells for most values of group_id, but is not present for one group_id, no model is fit for this cluster. Where clusters are strongly differentially abundant between conditions, this does happen in practice (in my data, this has resulted in ~25% of clusters returning no results, instead of returning partial results).
It should be possible to reduce the design matrix appropriately and rerun these checks, then run pbDS and add NA values for the excluded parameters.
pbDSremoves samples not satisfyingmin_cells/ outliers requirements. The checks inpbDS.R:147-149check that after these exclusions, it is still possible to fit the full model.This means that where a cluster has many cells for most values of
group_id, but is not present for onegroup_id, no model is fit for this cluster. Where clusters are strongly differentially abundant between conditions, this does happen in practice (in my data, this has resulted in ~25% of clusters returning no results, instead of returning partial results).It should be possible to reduce the design matrix appropriately and rerun these checks, then run
pbDSand addNAvalues for the excluded parameters.