vitkl
commented
1 year ago Is it statistically appropriate to conduct an enrichment analysis (e.g. wilcoxon rank sum test) of each cluster using this metrics, and assign cell types based on the results?
I don't fully understand your problem. If you want to find cell types enriched in a particular anatomical area you can use any statistical test (wilcoxon, t-test).
If you want to define areas using cell abundance you can binarise the cell abundance. For example, I use 99.9 quantiles (or top-{10-50} locations) to get locations most strongly enriched in the cell type of interest. Then you can use the binary labels to compute to perform enrichment tests wrt other info - such as other cell types, and anatomical features.
It is incorrect to label Visium locations with 1 cell type label because each location contains many cells cell types.
jeji0209
Please use the template below to post a question to https://discourse.scverse.org/c/ecosytem/cell2location/.
N_cells_per_locationanddetection_alpha.batch_keyfor reference NB regression.Hello, thank you for development of this tool. My goal is to subset a group of spots that is estimated to be a particular region, and conduct further downstream analysis within that region, e.g. DGE analysis between different conditions. I have conducted clustering via Scanpy, and one cluster indicates my region of interest, based on biological ground truth. However, for a concrete analysis of cell type annotation, I would like to utilize the 'q05_cell_abundance_w_sf' metrics in cell2location. Is it statistically appropriate to conduct an enrichment analysis (e.g. wilcoxon rank sum test) of each cluster using this metrics, and assign cell types based on the results?