[DeSeq2] some values in assay are negative

Question

[pentium3-user]

Hi, I'm trying to use DeSeq2 to calculate gene expression levels.

My count table (all_counts) looks like this

gene cell1 cell2
1 ENSMUST00000000001 712   1613
2 ENSMUST00000000003   0      0
3 ENSMUST00000000010   2      3
4 ENSMUST00000000028   0      0
5 ENSMUST00000000033   0      0

I'm call DESeqDataSetFromMatrix

dds <- DESeqDataSetFromMatrix(countData = all_counts,
                              colData = data.frame(c('gene', 'cell1', 'cell2')))

But I'm always getting this error

Error in DESeqDataSet(se, design = design, ignoreRank) : 
some values in assay are negative

But there are no negative values. I verified this by

> any (!is.integer(all_counts[,2]))
[1] FALSE
> any (!is.integer(all_counts[,3]))
[1] FALSE 
> any (all_counts[,2] < 0)
[1] FALSE
> any (all_counts[,3] < 0)
[1] FALSE

Any ideas what I'm doing wrong?

Answer 1

[zx8754]

From the manuals the countData must be a numeric matrix, from your example all_counts, it looks like a dataframe.

Try to convert it to matrix:

countDataMatrix <- as.matrix(all_counts[ , -1])

And keep the names of the genes in the rownames (as suggested by @DevonRyan in the comments):

rownames(countDataMatrix) <- all_counts[ , 1]

[pentium3-user]

Thx for your reply. I tried changing the call to

dds <- DESeqDataSetFromMatrix(countData = as.matrix(all_counts),
                          colData = data.frame(c('gene', 'cell1', 'cell2')))

But I'm still getting the same error message.

[zx8754]

Try dropping the first column: countData = as.matrix(all_counts[ , -1]

[pentium3-user]

Thank you very much, this works. So do I have to store the gene ids by my self in an extra vector or is there a better way how to keep the gene ids associated with the rows?

In this example they don't have to drop the gene ids but I don't know why it's not working in my case: https://informatics.fas.harvard.edu/differential-expression-with-deseq2.html

[Devon Ryan]

Set the gene IDs to be the row names of the matrix.

Answer 2

[Oliver]

I had the same problem, but also needed to set the column names explicitely and set the matrix content to integers. So I also want share what worked for me:

ctsTable <- read.table(ctsFile_path, sep="\t", header=FALSE, stringsAsFactors=FALSE)
countDataMatrix <- as.matrix(ctsTable[-1,-1])   # Extract counts & Ignore first column (geneID) and first row (sample names)
mode(countDataMatrix) <- "integer"   # Convert to integer
rownames(countDataMatrix) <- ctsTable[-1,1]     # Set rownames to geneIDs
colnames(countDataMatrix) <- ctsTable[1,-1]      # Set colnames to sample names

dds <-  DESeqDataSetFromMatrix(countData = countDataMatrix)