DESeq error

[clemen]

hi,

i have a problem, is my first approach R,
and when i want count my data....


> cds <- newCountDataSet(countsTable, condiciones)
Error in newCountDataSet(countsTable, condiciones) :
The countData is not integer.


and...


> as.integer(countsTable)
Error: (list) object cannot be coerced to type 'integer'


O_O i dont have idea!!!

please guide, thanks!!!

[Simon Anders]

Please tell us the output of "str(countsTable)"

[clemen]

Please tell us the output of "str(countsTable)"

thanks

the output

> str(countsTable)
'data.frame': 12555 obs. of 3 variables:
$ poli_control: num 1222.92 2.58 353.46 78.69 25.8 ...
$ poli_6hr : int 635 0 155 766 19 94 17 4 388 554 ...
$ poli_9hr : num 100.1 0 88.7 50 187.3 ...

[Simon Anders]

These are not count values. The input to DESeq must be read counts, i.e., integer numbers which tell you how many reads map to each gene in a sample.

[anurupa]

hii, i have prepared data as given in tutorial however wen i am giving this command it is showing the error
cds <- newCountDataSet( over, conds )

here is the error: Error in round(countData) : Non-numeric argument to mathematical function. can any one help me out

[Simon Anders]

Please post the output of 'str(over)' and 'str(conds)'.

[neha]

Difficulty in DESeq

Hello Simon,

I am quite new to R. I am trying to do differential expression analysis for RNA-seq data using DESeq. my original file looks like this:

wildtype mutant
A1BG 6 8
A1BG-AS1 2 1
A1CF 18 11
A2LD1 55 91
A2M 2 2
A2ML1 7 6
A4GALT 77 55
A4GNT 3 4
AAA1 8 7
AAAS 633.003 654

I am following the instructions given in DESeq tutorial

I am getting the following error
cds <- newCountDataSet(datafile, conds)
Error in newCountDataSet(datafile, conds) : The countData is not integer.

I tried to search several post and came across that one should use this command
datafile <- read.table(your_file, sep=',', header=FALSE, row.names=1) so that my first column will be treated as row but this again gives me error that line 1 does not contain three columns.

Can you please advice me what I am doing wrong.

Thanks in Advance
Neha

[Simon Anders]

Well, your count data is not integer. In the excerpt from the file you posted, the last line reads "AAAS 633.003 654". Where does the ".003" come from?

[Simon Anders]

And: in your 'read.table' command, you need to set "header=TRUE", because your file does have a header (i.e., the first line gives the column names).

[neha]

I got the read counts from Partek software. It does give read count in fraction.

[Simon Anders]

How can a read count be in fractions? What DESeq needs is the number of reads that map to each gene in each sample. A read either maps to a gene or it does not; hence, there cannot be fractional numbers. Whatever it is that Partek gives you, it does not seem to be read counts.

(In case Partek spreads reads that cannot be unambiguously mapped to several genes, adding fractions to each gene: Such data should not be used for differential expression analysis; we had a couple of threads here discussing this.)

[Jean]

(In case Partek spreads reads that cannot be unambiguously mapped to several genes, adding fractions to each gene: Such data should not be used for differential expression analysis; we had a couple of threads here discussing this.)

Hi Simon,
Sorry to hijack this but I can't find the posts referring to this. How do you recommend dealing with equally multiply-mapped reads? Count only unique maps and discard the others?

[Simon Anders]

See e.g. post #4 in this thread: http://seqanswers.com/forums/showthread.php?t=9129

[am@i]

hi
I am performing DE analysis using DEseq. In a DE analysis suppose the control has a read count value zero (0) and treatment has some number (=>1) then the DE will always be infinity. In such a case how to analyse it statistically like fold change, FDR etc

[Simon Anders]

In a naive calculation, such zeroes cause infinite fold changes. For the estimation of significance, this has never been an issue, because the fold change estimate does not directly enter into the p value calculation, i.e., it is no problem at all to get valid DE calls in the presence of zeroes.

For the interpretation of effect size, the infinities cause difficulties, which can be avoided by using shrinkage estimation. Please see our new paper on DESeq2 for a thorough discussion of the latter issue.

Moderated estimation of fold change and dispersion for RNA-Seq data with DESeq2
Michael I Love, Wolfgang Huber, Simon Anders
bioRxiv, 2014, doi: 10.1101/002832