Separate healthy from neoplastic cells using 5'-BCR seq.

[cv55]

Hello, thanks for the great tool! Our main goal is to separate cancer from healthy cells in our samples. We expect cancer cells to form clusters that consist of one clone and healthy cells to form clusters that encompass multiple clones. In one sample I get this picture:

How safe would it be to say the cluster that has many no_contig cells is the healthy one?
I know that these are barcodes that do not match between the GEX and BCR assays, but it seems to me that the big cluster is a unique clone.

Do you have any suggestions on how to separate cancer and healthy cells using Dandelion?

[zktuong]

Hi @cv55 , thanks for the interest in Dandelion!

Is your cancer a B-cell neoplasm and are those cells with no contigs B cells? Also it seems to me that some of the colours are overlapping. How does it look if the colours are more distinct?

I think it also depends on how you set up your umap here - did you remove VDJ genes from the highly variable features before you scale/regress/pca/neighbors/umap? if not, maybe remove them first and try again.

You might want to consider orthogonal ways of supporting this observation e.g. infercnvpy to examine if those no contig cells indeed have little/no CNV events