Batch correction in TNBC epithelial cells

Question

[TNF_alpha]

I applied harmony to my TNBC sc-rna seq data to correct batch effect and did the cell type annotation. Then I sub-setted the epithelial cells and re-clustered them . As the epithelial cells showing batch effect I applied harmony to them.

NormalizeData, FindVariableFeatures, ScaleData, RunPCA, RunHarmony, FindNeighbors, FindClusters, RunUMAP - used these Seurat functions on sub-setted epithelial cells.

(1) Is my pipeline correct? Especially re-applying the harmony to the sub-setted the TNBC epithelial cells.

(2) Without harmony: my epithelial cells(both normal & cancer) were showing batch effect but immune & stromal cells had cells from different patients.

With harmony: there are mixing of cells in the cancer epithelial cells as well though not highly mixed like other stromal & immue cells.

is this observation biologically correct ?

Answer 1

[ATpoint]

(1) Is my pipeline correct? Especially re-applying the harmony to the sub-setted the TNBC epithelial cells.

You can do that. No problem with this I would say in general. The idea is always that integration tries to group cells by underlying cellular identity rather than condition. So overlap of normal and cancer could occur, given it's the same cell of origin I would say. After all, it's only used for dimensionality reduction, so clustering, UMAP etc. Differential expression is done on the unintegrated data.

is this observation biologically correct ?

Nobody here can tell you, that's down to you finding out given your knowledge of the system and literature.