Even with the robust cutoffs that we applied to this dataset to remove likely low quality or doublet cells; we found a residual subset of the data that contain these cells. Upon clustering, doublet cells tended to project into the UMAP space as long streaks between two well-defined cell types. Low-quality cell types would project into the UMAP space as amorphous cell types without clear boundaries. Using these embedding features, we selected these clusters with Seurat’s FindClusters(resolution = 1) function, confirmed that they have the indicative QC metrics, and removed them from analyses.
