Does ‘Size factors should be positive’ error matter?

[Yiwei Niu]

Hi, I found several samples with the error "Error in calculating norm factors:Error in .local(x, ...): size factors should be positive", but scDblFinder still successfully outputed the results. I was wondering if this error affected the result. Thanks!

[Pierre-Luc]

Can you provide a traceback() of the error?

[Yiwei Niu]

Hey, thanks for your quick reply.

Here is the warning message. It's just warnings, and scDblFinder could run successfully.

Warning messages:
1: In .checkSCE(sce) :
  Some cells in `sce` have an extremely low read counts; note that these could trigger errors and might best be filtered out
2: In value[[3L]](cond) :
  Error in calculating norm factors:Error in .local(x, ...): size factors should be positive

I converted the warnings to errors and here is the output of traceback()

> traceback()
8: doWithOneRestart(return(expr), restart)
7: withOneRestart(expr, restarts[[1L]])
6: withRestarts({
       .Internal(.signalCondition(simpleWarning(msg, call), msg, 
           call))
       .Internal(.dfltWarn(msg, call))
   }, muffleWarning = function() NULL)
5: .signalSimpleWarning("Some cells in `sce` have an extremely low read counts; note that these could trigger errors and might best be filtered out", 
       base::quote(.checkSCE(sce)))
4: warning("Some cells in `sce` have an extremely low read counts; note ", 
       "that these could trigger errors and might best be filtered out")
3: .checkSCE(sce)
2: scDblFinder(as.SingleCellExperiment(srat))
1: as.Seurat(scDblFinder(as.SingleCellExperiment(srat)))

[Pierre-Luc]

Thanks for the details.
Yes, it matters: the normalization error was turned into a warning for cases where it's unavoidable, but what this means is that for the purpose of finding doublets, the data will not be normalized. Typically this works surprisingly well (for reasons I won't get into), but it's not optimal.
In general I would strongly recommend that you filter out cells that have too few counts, these are clearly more trouble than they're worth.
Pierre-Luc

[Yiwei Niu]

Thanks very much for the explanations! It helps a lot.