T-cell recognized as tumor cells

[oandrefonseca]

I am testing SCEVAN for stratifying malignant cells in CD45 sorted dataset, i.e., up to a certain limit, I am confident about the cells' status (malignant versus normals). On that note, I noticed that CD45+ cells are classified as tumor cells.

After further investigation, I noticed that those are mostly immune cells (NK and T-cells). I suppose that it might be affected by VDJ gene recombinations. Have you faced this issue? Would it be fair to remove those genes from my gene counts? What would be the best protocol or work around it?

[ATPs]

similar question. Not only T cell.

cell_type	filtered	normal	tumor
B cell	624	1268	3869
T cell	9657	11313	30878
dendritic cell	1541	2679	688
endothelial cell	598	2151	329
epithelial cell	6707	6857	50738
erythrocyte	46	1	nan
fibroblast	1244	3076	584
macrophage	2047	5672	1532
mast cell	154	462	651
neutrophil	1047	41	17
plasma cell	63	489	114

[ATPs]

https://www.cell.com/cancer-cell/fulltext/S1535-6108(21)00497-9 I have data from this paper. Cells from normal tissues were annotated as tumor cell.

[AntonioDeFalco]

@oandrefonseca Thank you for the report. No we have not addressed this issue unfortunately, I wanted to ask you to better understand the issue if you are analysing 10x samples generally some misclassification may be due to the noisiness of these data or samples with very low tumor purity. Can you share the heatmap generated by SCEVAN? Thank you

@ATPs Thanks for the tip, since I see data from many aggregated samples in your table, I wanted to ask you if to analyse the dataset you mentioned did you analyse each sample individually with SCEVAN or did you analyse the entire matrix containing all samples? Thank you.

[ATPs]

@AntonioDeFalco Thank you for your reply. I think it is individual sample, since I used the function in this example.

http://htmlpreview.github.io/?https://github.com/AntonioDeFalco/SCEVAN/blob/main/vignettes/multiSamples.html

library(SCEVAN)
results <- SCEVAN::multiSampleComparisonClonalCN(listCountMtx, analysisName = "all", organism = "human" , par_cores = 20)

My code looks like:

library(SCEVAN)
library(Seurat)
alldata <- qs::qread("Combined_samples.qs")

alldata.list <- SplitObject(alldata, split.by = 'donor_id')
listCountMtx <- list()
for (donor_id in names(alldata.list)) listCountMtx[[donor_id]] <- alldata.list[[donor_id]]@assays$RNA@counts

rm(alldata, alldata.list)
gc()
results <- SCEVAN::multiSampleComparisonClonalCN(listCountMtx, analysisName = "all", organism = "human" , par_cores = 20)

# combine results
xdf.predict <- do.call(rbind, results[[1]])
write.csv(xdf.predict,'Combined_samples.20230619SCEVAN.tumor_cell_prediction.csv')

The cell types were annotated in the original file.