[Xi Wang]

Just figured out that it was an out-dated version of SeqGSEA. With the latest version, it works with vectors.

But just out of curiosity, if I have 5 samples for each treatment group, then there are in a total of 252 combinations of randomly assigning labels. Then does it make sense to run 1000 permutation?

You are right. Maybe you can decrease the permutation time to 250, or increase your samples size. If there are 6 samples per group, the random label assignment makes 924 combinations. It will not be a problem if you have > 6 samples per group.

[Xi Wang]

I finally noticed that I need at least five samples per group, so I guess the problem is that I only have two tumors of interest...?

Thanks anyway!

Thale

Sorry that the current version of SeqGSEA cannot work on less than 5 samples per group.

Xi

[Yvone]

Hello there,

I encounter this problem. As I am using the DE-only approach, I dont't have a RCS object. Any ideas why the function genpermuteMat() doesn't take a vector? Thanks in advance!

> label <- as.factor(c(rep(0,6), rep(1,6)))
> permMat <- genpermuteMat(label, times=100)
Error: is(RCS, "ReadCountSet") is not TRUE

Just figured out that it was an out-dated version of SeqGSEA. With the latest version, it works with vectors.

But just out of curiosity, if I have 5 samples for each treatment group, then there are in a total of 252 combinations of randomly assigning labels. Then does it make sense to run 1000 permutation?

[Yvone]

Hello there,

I encounter this problem. As I am using the DE-only approach, I dont't have a RCS object. Any ideas why the function genpermuteMat() doesn't take a vector? Thanks in advance!

> label <- as.factor(c(rep(0,6), rep(1,6)))
> permMat <- genpermuteMat(label, times=100)
Error: is(RCS, "ReadCountSet") is not TRUE

[thaleko]

I finally noticed that I need at least five samples per group, so I guess the problem is that I only have two tumors of interest...?

Thanks anyway!

Thale

[thaleko]

Hi,

I have recently started using SeqGSEA to analyse RNAseq data from 12 tumors. (Paired-end, 70-90 mill reads, mapped with tophat, counted with htseq-count).

I would like to compare two of these tumors (samples v6 and v11, see below) to the other ten regarding certain gene sets. However, I run into the following issue when following 6.3 in the manual (DE only):

Genes with read count 0 across all 12 samples have been removed from my counts table.

> nrow(counts)
[1] 45418
> head(counts)
v1 v2 v3 v4 v5 v6 v7 v8 v9 v10 v11 v12
ENSG00000000003 43888 15516 15307 4287 17635 3342 3904 7235 22298 12464 4476 10353
ENSG00000000005 2238 35 27 23 81 0 12 46 335 39 4 56
ENSG00000000419 2781 3032 1805 2644 2027 2651 3383 2230 2416 2070 2503 1381
ENSG00000000457 1212 727 699 962 707 626 1054 944 950 647 507 493
ENSG00000000460 524 538 396 640 415 1467 852 1073 568 419 409 399
ENSG00000000938 173 1349 616 467 280 1769 987 1159 415 448 1591 679
> label <- as.factor(c(0,0,0,0,0,1,0,0,0,0,1,0))
> DEG <- newCountDataSet(counts,label)
> DEG <- estimateSizeFactors(DEG)
> DEG <- estimateDispersions(DEG, method="pooled", fitType="local")
> DEGres <- DENBStat4GSEA(DEG)
> permuteMat <- genpermuteMat(label, times=perm.times)
> DEpermNBstat <- DENBStatPermut4GSEA(DEG, permuteMat)
Error in { :
task 1 failed - "Parametric dispersion fit failed. Try a local fit and/or a pooled estimation. (See '?estimateDispersions')"


I am using estimateDispersions and estimateSizeFactors instead of runDESeq in order to use fitType=local and method=pooled, but am still experiencing trouble. Any idea what I'm doing wrong? Anyone else having this problem? Any help would be greatly appreciated!


Cheers,
Thale

[Xi Wang]

Thanks for your quick reply.

I have other questions.
1). Regarding independent filtering of weakly expressed genes in DE analyses, is it also needed in SeqGSEA?
2). Can I apply multi-factor design in SeqGSEA? In DESeq, I applied glm models to correct variations by other factoers than my interest, for example, calculate DE genes for the treatment while correcting for sex, batch effect. Does SeqGSEA also allow a multi-factor design?

1) I suggestion you feed all genes to SeqGSEA, and SeqGSEA would apply itself filter to exclude unexpressed genes.
2) Sorry with the current version cannot.

[Yvone]

Thanks for your quick reply.

I have other questions.
1). Regarding independent filtering of weakly expressed genes in DE analyses, is it also needed in SeqGSEA?
2). Can I apply multi-factor design in SeqGSEA? In DESeq, I applied glm models to correct variations by other factoers than my interest, for example, calculate DE genes for the treatment while correcting for sex, batch effect. Does SeqGSEA also allow a multi-factor design?

[Xi Wang]

Hi,

I came across the SeqGSEA package for GSEA of RNA-seq data.
We extensively work on cancer data sets and found this tool to be quite intriguing.

I was wondering if there is sample data set available to understand all the formats of the file. I am already have the DESEQ count files available.

~Thanks,
Rini

A set of example data can be found by executing the following command:
system.file("extdata", package="SeqGSEA", mustWork=TRUE)

[Xi Wang]

I noticed that SeqGSEA uses the object CountDataSet from DESeq. As I am currently using DESeq2, the main object for analyses are DESeqDataSet and cannot be adopted directively in SeqGSEA. I guess these two things are similar, but is there an easy way to transform DESeqDataSet to CountDataSet?

I will try to move to DESeq2 in the future but at present the most easy way to deal with this issue is just rerun DESeq analysis using the SeqGSEA pipeline. We have already provided one all-in command to go through all analyses.

Cheers
Xi

[rpauly]

Sample data available?

Hi,

I came across the SeqGSEA package for GSEA of RNA-seq data.
We extensively work on cancer data sets and found this tool to be quite intriguing.

I was wondering if there is sample data set available to understand all the formats of the file. I am already have the DESEQ count files available.

~Thanks,
Rini

[Yvone]

I noticed that SeqGSEA uses the object CountDataSet from DESeq. As I am currently using DESeq2, the main object for analyses are DESeqDataSet and cannot be adopted directively in SeqGSEA. I guess these two things are similar, but is there an easy way to transform DESeqDataSet to CountDataSet?

[Xi Wang]

Just wondering, what is a geneset file?
How do you define the geneset file?
Any example format?
Thank you.

Thanks for considering SeqGSEA.

The gene sets can be downloaded from http://www.broadinstitute.org/gsea/msigdb/index.jsp

Or in GMT format if you want to specify yourself
http://www.broadinstitute.org/cancer..._.28.2A.gmt.29

[wilson90]

Just wondering, what is a geneset file?
How do you define the geneset file?
Any example format?
Thank you.

[Xi Wang]

The latest vignette claims one can use easyRNAseq for read counts. Using easyRNAseq, should one count exons, or genes? Also, how should the file be outputted, as a summarized experiment? A count table? Normalized values for DESeq?

Thanks for checking the latest version of SeqGSEA :-)

You should also count reads on exons, and output the read counts per sample. As to the format, please refer to the files in the directory:

dat.dir = system.file("extdata", package="SeqGSEA", mustWork=TRUE)

Perhaps I will provide an easier way to connect easyRNAseq and SeqGSEA in the upcoming versions.

Cheers
Xi