Error using feature counts with DESeq2 : Cannot create DESeqData object

Question

[VBer]

I have a file of read counts with which I want to find the differentially expressed genes from using DeSeq2. The file was generated using feature counts.

           C11  C15 C19 C23 N11 N15 N19 N23

NM_000014   4422    14216   8885    17031   8162    4811    12536   8273
NM_000015   3         0 7   2   0   9   2   6
NM_000016   1063    1192    1608    1345    1118    951 943 1120
NM_000017   164     424 463 507 603 692 494 653
NM_000018   5193    12982   11382   11716   10030   14180   9379    13316
NM_000019   654     1103    1106    1184    743 497 569 844

When I try to create a DeSeq2 object using the DESeqDataSetFromMatrix I get the following error:

DESeq.ds <- DESeqDataSetFromMatrix(countData = readcounts,colData = sample_info,design = ~ condition)
Error in DESeqDataSet(se, design = design, ignoreRank): some values in assay are not integers
Traceback:

1. DESeqDataSetFromMatrix(countData = readcounts, colData = sample_info,  design = ~condition)
2. DESeqDataSet(se, design = design, ignoreRank)
3. stop("some values in assay are not integers")

I checked the entire read counts file, there are no integers in it, so I don't understand why this error keeps occuring. I tried running the sapply(readcounts, class) command as suggested in this thread (which did not give a clear solution) and get the following output:

C11 'numeric'
C15 'numeric'
C19'numeric'
C23 'numeric'
N11 'numeric'
N15 'numeric'
N19 'numeric'
N23 'numeric'

I tried using DESeqDataSet instead, but that requires a RangedSummarizedExperiment object from the function summarizeOverlaps from the Genomic Alignments package. The summarizeOverlaps function does the same job as featurecounts - generate read counts. I don't want to repeat that step.

[ATpoint]

How did you obtain the count matrix? Are these raw (=non-normalized) counts? Please show head(readcounts).

You can also try as.integer(readcounts) given that there are indeed integers but they are somewhat misclassified as characters of fators. How did you import the count matrix?

[VBer]

The file was given to me by my professor as a csv file. Yes these are raw counts. I imported the file using the read.table command. as.integer(readcounts) doesn't work because readcounts is a dataframe.

head(readcounts)
C11 C15 C19 C23 N11 N15 N19 N23
NM_000014   4422    14216   8885    17031   8162    4811    12536   8273
NM_000015   3   0   7   2   0   9   2   6
NM_000016   1063    1192    1608    1345    1118    951 943 1120
NM_000017   164     424 463 507 603 692 494 653
NM_000018   5193    12982   11382   11716   10030   14180   9379    13316
NM_000019   654     1103    1106    1184    743 497 569 844

[ATpoint]

Come on, as.integer(as.matrix(readcounts))

[VBer]

Ah sorry XP New to R.

Edit: Sorry it worked. Should I convert the integer object back to a data frame? Again, sorry if the question is too obvious.

readcounts <- as.integer(as.matrix(readcounts))
head(readcounts)
4422 3 1063 164 5193 654 
DESeq.ds <- DESeqDataSetFromMatrix(countData = readcounts,colData = sample_info,design = ~ condition)
Error in validObject(.Object): invalid class “SummarizedExperiment” object: nb of cols in 'assay' (1) must equal nb of rows in 'colData' (8)

[ATpoint]

Ah sorry XP New to R.

No problem, sorry did not intend to sound harsh :)

mode(readcounts) <- "integer" is the last thing I could think of.

[swbarnes2]

Have you tried importing just the top of the file? Maybe there is one line that got corrupted. Is dim(readcounts) what you expect? You might have a weird whitespace hiding in there somewhere.

[ATpoint]

tmp <- gsub(" ", "", readcount); mode(tmp) <- "integer"

might be worth a try

[VBer]

Hey, thank you SO MUCH for asking me to check the dimensions of my data frame. I was indeed missing around 8000 genes. I figured out the error; I was using this code to make my Gene_IDs unique and convert them to row names:

readcounts <- aggregate(readcounts[-1],readcounts[1],mean)
row.names(readcounts) <- readcounts[,1]
readcounts$Gene_ID <- NULL
head(readcounts)

That block of code was somehow chopping off the last 8k of my genes. I also figured that it was adding some non-integer values. So I replaced the code with this instead:

rownames(readcounts) = make.names(readcounts$Gene_ID, unique=TRUE) #To make unique rownames
readcounts$Gene_ID <- NULL
head(readcounts)
                   C11  C15 C19 C23 N11 N15 N19 N23
NR_075077   0   5   0   4   8   2   2   6
NM_001276352    0   5   0   4   8   2   2   6
NM_001276351    0   5   0   4   8   2   2   6
NM_000299   1220    5980    5089    2792    5223    9731    4365    4755
NM_001005337    1220    5980    5089    2792    5223    9731    4365    4755
NM_012102   7436    15741   13205   14024   15995   14659   9167    12504

I found that bit of magic here. I was able to create a DESeqObject now. Thanks swbarnes and ATpoint!