Hello,
I am analyzing a public scRNA dataset using Seurat. My goal is to observe variation inside individual cell clusters according to a condition (e.g. the diet) in a visual way using a dimensional reduction plot, e.g. observing sub-clusters of cells that spatially aggregate according to this condition. My initial idea was to subset the dataset for every cell cluster and run DimPlot()
with the UMAP reduction (that was used to cluster and annotate all the cell clusters for the whole dataset) with group.by = 'condition'
and highlighting the cells by every condition. I attach an example image for a single cluster in which can be observed that for instance the blue cells appear on the left while the purple appear on the right.
However I got some questions during the process.
1- Is it OK to use this UMAP reduction for this purpose?
2- Would it be better to run the standard PCA pipeline again (NormalizeData() + FindVariableFeatures() + ScaleData() + RunPCA()
) using separately a subset dataset for every cluster? I wonder this in case the dim. reduction techniques applied before to cluster cell types were not that sensitive for the diet condition but instead they were capturing differences for the cell types.
Thank you for the help.