Error in design of voom object

[kelshmo]

I'm running varianceParition 1.19.6. To repro the error from voomWithDreamWeights:

data(varPartDEdata)
countMatrix <- countMatrix[1:1000,]
gene_expression <- edgeR::DGEList(countMatrix)
gene_expression <- edgeR::calcNormFactors(gene_expression)
voom_gene_expression <- variancePartition::voomWithDreamWeights(counts = gene_expression$counts,
                                                                formula = "~ Disease + (1 |DiseaseSubtype)",
                                                                data = metadata,
                                                                quiet = TRUE)
voom_gene_expression$design

Error:

Error in x[[what]] : object of type 'closure' is not subsettable

[GabrielHoffman]

I will this this tomorrow, but in the meantime the issue is due to the fact that your formula is really a string. If you remove the quotes it should work

[GabrielHoffman]

Fixed in v1.19.7. Confirm?

[kelshmo]

The outcome is the same with both an unquoted and a quoted formula.

> voom_gene_expression$design
standardGeneric for "design" defined from package "BiocGenerics"

function (object, ...) 
standardGeneric("design")
<bytecode: 0x7fbbadbae4a8>
<environment: 0x7fbbadbadac0>
Methods may be defined for arguments: object
Use  showMethods("design")  for currently available ones.

$design
Error in x[[what]] : object of type 'closure' is not subsettable

I'm working in R 4.0.2.

[kelshmo]

This error is still a problem in 1.19.8. Here is another minimum reproducible example:

metadata <- tibble::tribble(
  ~samples, ~sex, ~batch, ~age, ~RIN, ~diagnosis,
  "S567", "F", "1", "68", "7.5", "dx",
  "S453", "F", "2", "60", "7", "dx",
  "S231", "M", "2", "54", "7", "dx",
  "S444", "M", "2", "55", "7", "dx",
  "S555", "F", "1", "68", "7.5", "dx",
  "S304", "F", "2", "60", "7", "ct",
  "S232", "M", "2", "54", "7.5", "ct",
  "S447", "M", "2", "55", "7.5", "ct"
)

metadata <- tibble::column_to_rownames(metadata, var = "samples")

counts <- tibble::tribble(
  ~geneId, ~S567, ~S453,~S231, ~S444, ~S555, ~S304, ~S232, ~S447,
  "ENSG1", 0, 0, 500, 0, 10, 43, 432, 45,
  "ENSG2", 23, 25, 29, 30, 56, 89, 73, 70,
  "ENSG3", 84, 82, 0, 0, 98, 80, 200, 201,
  "ENSG4", 0, 204, 350, 790, 353, 456, 555, 890,
  "ENSG5", 0, 0, 0, 70, 0, 0, 0, 70
)

counts <- tibble::column_to_rownames(counts, var = "geneId")

counts <- as.matrix(counts)

gene_expression <- edgeR::DGEList(counts)
gene_expression <- edgeR::calcNormFactors(counts)
voom_gene_expression <- variancePartition::voomWithDreamWeights(counts = gene_expression,
                                                                formula = "~ diagnosis + (1|batch)",
                                                                data = metadata,
                                                                quiet = TRUE)
voom_gene_expression$design

[kelshmo]

In 19.9.9 I'm returning a new error using the above code chunk:

The gene_expression object is length 8 and the colnames match the rownames of metadata and is also length 8.

> voom_gene_expression <- variancePartition::voomWithDreamWeights(counts = gene_expression,
+                                                                 formula = "~ diagnosis + (1|batch)",
+                                                                 data = metadata,
+                                                                 quiet = TRUE)
Error in .fitVarPartModel(exprObj, formula, data, REML = REML, useWeights = useWeights,  : 
  the number of samples in exprObj (i.e. cols) must be the same as in data (i.e rows)

[GabrielHoffman]

This error is intended, but not that informative. I have updated it to read:
counts is type 'numeric' and can't be converted to matrix unambiguously

The issue is that that counts you passed was a vector and there is no way to tell if you mean that as the row or a column of a matrix. So it throws the new error in v1.19.10

metadata <- tibble::tribble(
  ~samples, ~sex, ~batch, ~age, ~RIN, ~diagnosis,
  "S567", "F", "1", "68", "7.5", "dx",
  "S453", "F", "2", "60", "7", "dx",
  "S231", "M", "2", "54", "7", "dx",
  "S444", "M", "2", "55", "7", "dx",
  "S555", "F", "1", "68", "7.5", "dx",
  "S304", "F", "2", "60", "7", "ct",
  "S232", "M", "2", "54", "7.5", "ct",
  "S447", "M", "2", "55", "7.5", "ct"
)

metadata <- tibble::column_to_rownames(metadata, var = "samples")

counts <- tibble::tribble(
  ~geneId, ~S567, ~S453,~S231, ~S444, ~S555, ~S304, ~S232, ~S447,
  "ENSG1", 0, 0, 500, 0, 10, 43, 432, 45,
  "ENSG2", 23, 25, 29, 30, 56, 89, 73, 70,
  "ENSG3", 84, 82, 0, 0, 98, 80, 200, 201,
  "ENSG4", 0, 204, 350, 790, 353, 456, 555, 890,
  "ENSG5", 0, 0, 0, 70, 0, 0, 0, 70
)

counts <- tibble::column_to_rownames(counts, var = "geneId")

counts <- as.matrix(counts)

# code from before: running calcNormFactors on counts directly is not what you want.
# gene_expression <- edgeR::DGEList(counts)
# gene_expression <- edgeR::calcNormFactors(counts)

# correct code that avoids this error
DGE <- edgeR::DGEList(counts)
gene_expression <- edgeR::calcNormFactors(DGE)

form = ~ diagnosis + (1|batch)

voom_gene_expression <- variancePartition::voomWithDreamWeights(counts = gene_expression,
                                                                formula = form,
                                                                data = metadata,
                                                                quiet = TRUE)

voom_gene_expression$design

[kelshmo]

Oops that was a typo! Thanks for updating the error message.

However, when I copy and paste the above I still get the error:

Error in variancePartition::voomWithDreamWeights(counts = gene_expression,  : 
  object 'design' not found

Running varianceParition 1.19.10. Since that is a limma error (I think) limma_3.44.3 is in the namespace but not attached to the session.

[GabrielHoffman]

Wow, this is a mysterious one! Can you send traceback() and sessionInfo()

Also are you running in a fresh R session?

[kelshmo]

This works in a fresh R session. 💯