DESeq2 column/row arrangement

[bioBob]

Hi,

I am trying to automate some of my analysis and am pulling in condition and data information from a master experiment design sheet and a HTSeq count table for the entire experiment. Because I am getting the sheet from the investigator and the HTSeq count is in run order using identifiers from the lab, the two sheets might be in different orders. Example:

slim_expt[,1:2]
       SampleID     Day
P3-D1     P3-D1 control
P3-D2     P3-D2 control
P3-D3     P3-D3 control
P3-D4     P3-D4 control
P12-D1   P12-D1 treated
P12-D2   P12-D2 treated
P12-D3   P12-D3 treated
P12-D4   P12-D4 treated

head(slim_HTS_table)
                   P12-D1 P12-D2 P12-D3 P12-D4 P3-D1 P3-D2 P3-D3 P3-D4
ENSMUSG00000000001   1546   1322   1248   1156  2162  2211  2811  2223
ENSMUSG00000000003      0      0      0      0     0     0     0     0
ENSMUSG00000000028     14     23     11     17    42    30    53    55

Note that the first sample in the experiment design sheet is not the first sample on the HTSeq count matrix.

Now I do this:
slim_dds<-DESeqDataSetFromMatrix(countData = slim_HTS_table, colData = slim_expt, design = ~Day)

And wallah, the sample names are now in order, but the count matrix order was preserved, ie samples are now misassigned.

head(counts(slim_dds))
                   P3-D1 P3-D2 P3-D3 P3-D4 P12-D1 P12-D2 P12-D3 P12-D4
ENSMUSG00000000001  1546  1322  1248  1156   2162   2211   2811   2223
ENSMUSG00000000003     0     0     0     0      0      0      0      0
ENSMUSG00000000028    14    23    11    17     42     30     53     55

Is there an easy flag to fix this or do I simply need to sort both data frames (by row or column as appropriate) prior to creating the DESeq data set?

[Simon Anders]

Yes, you need to bring your count table into the right order beforehands.

Do, for example:

slim_dds <- DESeqDataSetFromMatrix(
countData = slim_HTS_table[ , rownames(colData) ],
colData = slim_expt,
design = ~Day)

However, it is unfortunate that DESeqDataSetFromMatrix silently mixed up the column names. It takes the rownames from the colData frame and used them as sample names, overwriting the columns names of the count matrix. It should issue a warning, or even an error, in such case. We'll change that,

[Michael Love]

Good idea, I will implement a check for this in the devel branch. I think it should be an error.

[Michael Love]

I've implemented a test for this in version 1.3.46. If the names are the same, but in the wrong order, it throws an error.

> count <- matrix(1:20, ncol=4)
> colnames(count) <- c("a","b","c","d")
> cols <- data.frame(condition=factor(c(1,1,2,2)))
> rownames(cols) <- c("1","2","3","4")
> DESeqDataSetFromMatrix(count, cols, ~ condition)
class: DESeqDataSet
dim: 5 4
exptData(0):
assays(1): counts
rownames: NULL
rowData metadata column names(0):
colnames(4): 1 2 3 4
colData names(1): condition
> rownames(cols) <- c("a","b","c","d")
> DESeqDataSetFromMatrix(count, cols, ~ condition)
class: DESeqDataSet
dim: 5 4
exptData(0):
assays(1): counts
rownames: NULL
rowData metadata column names(0):
colnames(4): a b c d
colData names(1): condition
> rownames(cols) <- c("b","c","d","a")
!> DESeqDataSetFromMatrix(count, cols, ~ condition)
Error in DESeqDataSetFromMatrix(count, cols, ~condition) :
rownames of the colData:
b,c,d,a
are not in the same order as the colnames of the countData:
a,b,c,d