LTLA
commented
6 years ago This sounds more like an issue with scater's runTSNE rather than mnnCorrect, especially if you can successfully generate a PCA off the MNN-corrected values. Indeed, in the code above, you're not even running the t-SNE on the MNN-corrected values, but instead the "logcounts" assay.
I'm a bit bemused about how it is possible to have no dimensions in the svd function - that would suggest that you have no cells or no genes within the Rtsne call (itself within runTSNE). Off the top of my head, that would involve all-zero rows, or possibly a dataset full of duplicate cells.
I would suggest running debug(runTSNE), stepping through the function, and investigating the value of vals that is being passed into the Rtsne call. If this has zero dimensions, then it's a fault in runTSNE; if it has non-zero dimensions, then it's a fault in Rtsne.
In addition, here are some more general comments:
- The DropletUtils package can directly read 10X results into a
SingleCellExperimentwith theread10xCountsfunction, no need for a manual conversion. - I strongly recommend against using
@to get or set data values in a S4 class. This is widely considered to be Bad Practice; use the appropriate getters/setters instead. For example,assay(blah)andassay(blah) <- x. - Always report your session information when posting issues, see
sessionInfo(). When using Bioconductor packages, make sure you're using the latest version (BioC 3.7).
MwrulesC
Hi, I don't have much experience with scran, but I've recently been testing the use of mnnCorrect on some of the samples I have. Everything seems to work and I can generate figures up until I try to run TSNE calculations on the generated output. At that point I get the error message: Error in svd(x, nu = 0, nv = k) : a dimension is zero
After some digging around, I haven't quite figured out how to solve this error message in this particular instance. I was just wondering if you've had experience with something like this before, and if I should be doing some extra filtering or normalization pre/post analysis to avoid this.
The general overview of the code I use is the following:
rawData <- Read10X(filepath1) fixed <- SingleCellExperiment(assays = list(counts=as.matrix(rawData)), colData = list(Sample = rep('Fixed',dim(rawData)[2]))) fixed <- computeSumFactors(fixed) fixed <- normalize(fixed)
rawData <- Read10X(filepath2) mixed <- SingleCellExperiment(assays = list(counts=as.matrix(rawData)), colData = list(Sample = rep('Mixed',dim(rawData)[2]))) mixed <- computeSumFactors(mixed) mixed <- normalize(mixed)
out <- mnnCorrect(fixed@assays$data$logcounts, mixed@assays$data$logcounts)
mixed@assays$data$mnnCounts <- out$corrected[[2]] fixed@assays$data$mnnCounts <- out$corrected[[1]] combined <- cbind(fixed, mixed) combined <- runPCA(combined, exprs_values = "mnnCounts") plotReducedDim(combined, use_dimred = "PCA", colour_by = "Sample", shape_by = "Sample")
combined <- runTSNE(combined, exprs_values = "logcounts") plotTSNE(combined, colour_by = "Sample", shape_by = "Sample")
Thanks