Errors in MarkerBasedDecomposition

Question

Errors in MarkerBasedDecomposition

Closed this issue 5 years ago · 6 comments

Hi Brandon,

I want to use the BisqueRNA package to infer the cell types of my bulk data. I have my own markers and am using MarkerBasedDecomposition function. However, with different cutoffs of min_gene, I got different errors.

res <- MarkerBasedDecomposition(bulk.eset, markers = mymarker)
The error was "Clusters must have a minimum of 5 unique marker genes"

res <- MarkerBasedDecomposition(bulk.eset, markers = mymarker, min_gene = 1)
The error was "Error in base::eigen(varcov):infinite or missing values in 'x'"

res <- MarkerBasedDecomposition(bulk.eset, markers = mymarker, min_gene = 2)
The error was "Error in 1:ng: argument of length 0".

I attached the bulk data and markers I used.
exampleData.rda.zip

Please help me out.

Thanks,
Frank

Answer 1 · 2019-12-04T18:36:02.000Z

Hi Frank,

We diagnosed the bug and will push the fixes soon. The errors are due to a cell-type with only one marker gene. This should be avoided since the marker-based decomposition uses a PCA-based approach that does not apply to cases with only one gene. I'll let you know when the package is updated.

Thanks,
Brandon

Answer 2 · 2019-12-07T01:15:14.000Z

Package has been updated on GitHub. Let me know if there are still any issues. The update has not propagated to CRAN or Bioconda yet, so you'll need to manually update if you need it immediately

Answer 3 · 2019-12-09T17:32:34.000Z

Thank you for the updates. It works well now. Still have a quick question: what are the values of the bulk.props? My understanding is the weight of the similarity of a sample (bulk) and a cluster (single-cell). Correct me if I don't get it. Thanks, Frank

…

On Fri, Dec 6, 2019 at 8:15 PM Brandon Jew ***@***.***> wrote: Package has been updated on GitHub. Let me know if there are still any issues. The update has not propagated to CRAN or Bioconda yet, so you'll need to manually update if you need it immediately — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#7?email_source=notifications&email_token=AI2CASBG3PMMJT5253D34NLQXL2KHA5CNFSM4JU6HG62YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEGFZ7IY#issuecomment-562798499>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AI2CASE5UB34VZTPLHWXDNLQXL2KHANCNFSM4JU6HG6Q> .

Answer 4 · 2019-12-16T20:17:49.000Z

Hi Frank, the bulk.props contains (relative) cell type proportions estimates for each sample. These are principal component scores and so they do not represent true proportions. So samples with higher values for a cell type will have overall higher expression of that cell type's marker genes. Hope this make sense!

Thanks,
Marcus

Answer 5 · 2020-01-02T16:51:25.000Z

Hi Marcus, Clear explanation. However, in my case, some of the values are negative. Is it correct? Best, Frank

…

On Mon, Dec 16, 2019 at 3:17 PM marcalva ***@***.***> wrote: Hi Frank, the bulk.props contains (relative) cell type proportions estimates for each sample. These are principal component scores and so they do not represent true proportions. So samples with higher values for a cell type will have overall higher expression of that cell type's marker genes. Hope this make sense! Thanks, Marcus — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#7?email_source=notifications&email_token=AI2CASHVG6MHSNIITAF3O6DQY7O65A5CNFSM4JU6HG62YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEG76TSQ#issuecomment-566225354>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AI2CASCHMBP2JKVN47DMEGLQY7O65ANCNFSM4JU6HG6Q> .

Answer 6 · 2020-01-03T00:08:38.000Z

Hi Frank,

That is correct. Some values can certainly be negative.

Best,
Marcus