navinlabcode/CellTrek

Coembedding shows batch effect

Laolga opened this issue · 4 comments

Laolga commented

I obtained not very expected results from celltrek (some cell types are mapped randomly) and it made me wonder if something went wrong downstream. My main suspect is coembedding where data types are very well separated:
image
I followed the tutorial in the readme and did not apply any extra filtering/gene selection/preprocessing steps. Could you suggest what might be the issue?

williamhsy commented

Hi,
just wonder if the scRNA-seq and ST data are from same tissue sample and adjacent tissues? It would be great if you can provide more context here.

diyang1354 commented

Hi,
I'm facing the same issue. I used matched scRNA-seq and ST, but most of the cells were separated by sc and st.
Would this be an issue for downstream analysis?
image

williamhsy commented

I guess so. but there is still one blob on the umap showing mixture of sc and st. I wonder if it is because that the cell compositions are quite different between your sc and st data?

Laolga commented

Sorry for disappearing. In my case, the samples were from the same patient's tumor