Question about cell identity

[superSXB]

I have a question. When performing the identification of large classes of cells, the cell types identified solely by marker genes match the results integrated with the transcriptome. However, when it comes to sub - types, there is a significant mismatch. So, my question is, in this situation, should I trust the sub - types identified by marker genes or the results integrated with the transcriptome? Thanks a lot.

[superSXB]

Here is my proposed solution. I'm not sure if it's feasible. I plan to take the cell types obtained from the integrated transcriptome data as the main reference. For example, both the UMAP positions in the transcriptome data and the trajectory in Monocle3 show a sequence of A - B - C, while in scATAC data, it shows A - notB - C. I will then check the expression of B - cell markers in the gene activity of scATAC - seq data. If there is expression of these markers in the notB cells, I will modify the meta.data to classify notB as B cell type. Is this approach correct?

[kryptonitecircuit2002]

I am not sure what you mean by subtype. Assuming that you are talking about RNA and ATAC layers forming different cluster patterns, another method to work around is to name the celltypes of RNA, ATAC and WNN as seperate layers, especially if there is a case of two RNA clusters merging in ATAC. I do that and then make a heatmap of ARI of each cluster across RNA and ATAC to show the correspondence. Works for me most of the time.

Furthermore, while gene activity is a good methos, you can also check expression of your markers in your ATAC clusters using FeaturePlot(Seurat, features = "feature_name", reduction = "atac.umap"). Even though it uses counts from SCT layer to compute the gene expression and overlay it on ATAC clusters, so far, it has been accurate in my dataset.