Filter out doublet and Clustering

Question

[synat.keam]

Dear all, fellows,

I used DoubletFinder from Chris https://github.com/chris-mcginnis-ucsf/DoubletFinder to filter out doublet cells for each sample individually and used standard seurat pipeline to normalize the data as using following functions because I cannot use SCTransform with seurat 5 as DoubletFinder did not upgrade to match with Seurat5 yet.

NormalizeData(object = XXX)
FindVariableFeatures(object = XXX)
ScaleData(object = XXX)
RunPCA(object = XXX)
FindNeighbors(object = XXX, dims = 1:X)
FindClusters(object = XXX)
RunUMAP(object = XXX, dims = 1:X)

After that, I used DoubletFinder pipeline to filter out doublet cells and save singlet object for individual sample. Next, I merged all singlet objects for analysis. After merging, I used SCTransform to normalize the data, runPCA, integration using harmony and RunUMAP, FindNeighbor, and FindCluster etc.

My question, since I normalized data using seurat pipeline for doublet filtering and then switched to SCTransform after merging all singlet object for clustering. Is this the right way? The reason I sticked with SCTransform because I tended to get a good cluster with it. However, weird thing came up. Treg (Foxp3) population and CD8 population and CD4 population looked strange in Cluster 7, 8, 11, 13 as they have a bit of those three population. Generally, Treg should stay together in Cluster 6 rather spreading in different cluster far away for each other. Not sure whether these are also doublet. If so, they are big population. What could be wrong here and what else should I do. I am just stuck a bit. I attached the picture. Looking forward to some comments.

Regards,
Sk

[Picasa]

Did you find an answer about it ?

Answer 1

[jv]

It's a common misconception but you shouldn't judge the clustering results using UMAP reduction plots, the two are not the same thing.