[sindrle]

This is what I got with default setting (parametric), does it look ok or should I try mean or local fit?

One question:
What happens when I increase the "maxit" value? From 100 to 200, 159 was reduced to 110 rows that did not converge in beta. I dont know what this means.

Thanks!

Rplot.pdf

[Michael Love]

As there is not a clear dependence of the dispersion estimates on the mean of normalized counts, I would recommend to use fitType="mean".

[alisrpp]

The local fit usually does well as a substitute for the parametric fit, but I also warn the user that they should manually examine the plot generated by plotDispEsts(dds). You should check this plot to make sure the local fit is capturing the trend in the black points, as it should be.

This is my dispersion plot, is not exactly like in the manual... so i don't know what to think.
Can you give me your opinion, please?

Thanks,

[Michael Love]

You don't need to fix anything.

There are 3 levels of messages to the user in R: message, warning and stop/error. A warning is a non-fatal problem. In this case, I am warning the user that the default method for fitting the dispersion trend line didn't work so the software is doing something else instead.

The way to read the warning is that "algorithm did not converge" within the function estimateDispersionsFit. Then it says that the parametric fit failed, and that a local fit will be tried instead.

If you look up ?estimateDispersionsFit it will link over to ?estimateDispersions, here I describe the 3 options for the fitType argument, and the default is parametric. The local fit usually does well as a substitute for the parametric fit, but I also warn the user that they should manually examine the plot generated by plotDispEsts(dds). You should check this plot to make sure the local fit is capturing the trend in the black points, as it should be. If the local fit does not seem to capture the trend in the black points, you can use the last option fitType="mean" instead. Note that the fitType arguments can be provided to DESeq() as well as to estimateDispersions().

For changing columns names, check ?colnames.

[alisrpp]

Now i'm getting this error:

Warning messages:
1: glm.fit: algorithm did not converge 
2: In estimateDispersionsFit(object, fitType = fitType, quiet = quiet) :
  parametric fit failed, trying local fit. use plotDispEsts() to check the quality of the local fit

This is my code:

#Count matrix input
Cele_SPvsEB = read.csv (file.choose(), header=TRUE, row.names=1)
CeleDesign <- data.frame(
  row.names = colnames(Cele_SPvsEB$target_id),
  condition = factor(c("SP", "SP", "EB", "EB")))
dds <- DESeqDataSetFromMatrix(countData = Cele_SPvsEB,
                              colData = CeleDesign,
                              design = ~ condition)
dds

I obtained this:

class: DESeqDataSet 
dim: 198821 4 
exptData(0):
assays(1): counts
rownames(198821): comp100_c0_seq1 comp100004_c0_seq1 ... comp99994_c0_seq1 comp99999_c0_seq1
rowData metadata column names(0):
colnames(4): 1 2 3 4
colData names(1): condition

And after running:

#Differential expression analysis
dds <- DESeq(dds)

(...) the error warning is appearing.

Any idea how to fix that?

I also want to change the colnames, how can i do that?

Thanks.

[alisrpp]

Done! I updated R and now is working perfecly.
Thanks Michael!

[Michael Love]

Googling the setDummyField error I came to this thread

http://permalink.gmane.org/gmane.science.biology.informatics.conductor/48611

Where they suggest there might be a problem with using R 3.0.0 with packages built with R 3.0.1. Could this be the case? Can you paste your sessionInfo()

[alisrpp]

Sure!

> traceback()
14: get(name, envir = asNamespace(pkg), inherits = FALSE)
13: methods:::.setDummyField
12: (function (value) 
    {
        if (missing(value)) 
            `.->data`
        else {
            methods:::.setDummyField(.self, ".->data", "SimpleList", 
                "data", FALSE, value)
            value
        }
    })(quote(<S4 object of class "SimpleList">))
11: assign(field, value, envir = env)
10: envRefSetField(.Object, field, classDef, selfEnv, elements[[field]])
9: methods::initRefFields(.Object, classDef, selfEnv, list(...))
8: initialize(value, ...)
7: initialize(value, ...)
6: methods::new(def, ...)
5: .ShallowSimpleListAssays$new(data = assays)
4: .local(assays, ...)
3: SummarizedExperiment(assays = SimpleList(counts = countData), 
       colData = colData, ...)
2: SummarizedExperiment(assays = SimpleList(counts = countData), 
       colData = colData, ...)
1: DESeqDataSetFromMatrix(countData = Cele_SPvsEB, colData = CeleDesign, 
       design = ~condition)

> head(Cele_SPvsEB)
                    C4_APR10_spermatic_cysts CRL_2APR10_spermatic_cysts CRL_1_15JUL11_embryos
comp100_c0_seq1                           21                         24                     0
comp100004_c0_seq1                        48                          0                   494
comp1000074_c0_seq1                       42                          0                    55
comp1000109_c0_seq1                       32                         18                     0
comp100011_c0_seq1                        40                          3                   298
comp100013_c0_seq1                        92                          0                    62
                    CRL_2_15JUL11_embryos
comp100_c0_seq1                         0
comp100004_c0_seq1                      0
comp1000074_c0_seq1                     0
comp1000109_c0_seq1                     0
comp100011_c0_seq1                      0
comp100013_c0_seq1                      0

[Michael Love]

Can you show the output of traceback()

and also can you show the output of head(Cele_SPvsEB)

[alisrpp]

Hi,

I'm new in DESeq2 and i'm having problems to create my dds.

I have a matrix with eXpress results (read counts) with 2 conditions (SP, EB) and 2 replicates/condition.

This is what i'm doing:

Cele_SPvsEB = read.csv (file.choose(), header=TRUE, row.names=1)
CeleDesign <- data.frame(
  row.names = colnames(Cele_SPvsEB),
  condition = factor(c("SP", "SP", "EB", "EB")))
dds <- DESeqDataSetFromMatrix(countData = Cele_SPvsEB,
                              colData = CeleDesign,
                              design = ~ condition)
dds

And i'm getting this error message:

Error in get(name, envir = asNamespace(pkg), inherits = FALSE) : 
  object '.setDummyField' not found

Any help?
Thanks.

[Michael Love]

We generalized the code, and as these columns are only appropriate for a 2 condition analysis, they were dropped.

You can get these with some custom code though:

baseMeanA <- rowMeans(counts(dds,normalized=TRUE)[,colData(dds)$condition == "A"])

More general:

baseMeanPerLvl <- sapply(levels(colData(dds)$condition), function(lvl) rowMeans(counts(dds,normalized=TRUE)[,colData(dds)$condition == lvl]))

[KHubbard]

baseMeanA and baseMeanB

Hello,

In DESeq, it was possible to view/export the baseMean values for each condition (baseMeanA and baseMeanB) in addition to the overall baseMean. In DESeq2, I have only been able to view/export the overall baseMean. Is it possible to acquire the baseMean values for each condition as well? Thanks!

[Michael Love]

I committed this fix in version 1.1.31 of the devel branch. It made sense to move the outlier detection by Cook's distance as well as all p-value adjustment to the results() function.

[Michael Love]

You're right. Thanks for pointing this out.

I will work on a fix, probably moving the cooksCutoff argument to results() function, so it can be modified without rerunning the GLM fitting.

[NateP]

Glad to see contrasts appear in DESeq2, thanks!

One minor thing I noticed when running contrasts:
When you use the Cooks Outlier filtering on the base analysis, the outlier genes (with "NA" pvalues in the regular analysis) are no longer filtered when you do a contrast. I'm not sure if that is intentional or not?