Hello
I have a question regarding to the replicates for DESeq.
Here is my data design..
> head(data)
KYM_pre KYM_post GDY_pre GDY_post HKS_pre HKS_post KWK_pre KWK_post PYM_pre PYM_post SHM_pre SHM_post
A1BG 13 24 25 10 47 22 19 12 27 45 50 61
A2M 3 4 4 8 37 8 19 6 4 3 7 11
A2M-AS1 7 1 1 1 10 1 7 7 0 0 0 8
......
> meta
patient drug
KYM_pre 1 pre
KYM_post 1 post
GDY_pre 2 pre
GDY_post 2 post
HKS_pre 3 pre
HKS_post 3 post
KWK_pre 4 pre
KWK_post 4 post
PYM_pre 5 pre
PYM_post 5 post
SHM_pre 6 pre
SHM_post 6 post
=====================================
My goal is to find DEG responsing to the drug..
I tried two strategy neither of them worked..
1) Frankly, I don't have replicates.. so..
d_rare <- newCountDataSet(data, meta)
d_rare <- estimateSizeFactors(d_rare)
d_rare <- estimateDispersions(d_rare, method="blind", sharingMode="fit-only")
dh_fit1 = fitNbinomGLMs(d_rare, count ~ patient + drug)
dh_fit0 = fitNbinomGLMs(d_rare, count ~ patient)
Unfortunately, this model return NO DEG based on adjusted pvalue.
2) I assumes 6 patients having same disease are biological replicates for more solid estimation for dispersion
d_rare <- newCountDataSet(data, meta)
d_rare <- estimateSizeFactors(d_rare)
d_rare <- estimateDispersions(d_rare) # which means using default option (maximum likelihood estimation)
Again, in this second trial I got the following error
"None of your conditions is replicated. Use method='blind' to estimate across conditions, or 'pooled-CR', if you have crossed factors "
Does anyone have an idea? I have replications for patient(e.g. 1 1 2 2 3 3 44 ...etc) and drug(pre/post) too. why I am getting this message?? My suspection is that when I consider the cross design.. for example, for fixed factor(e.g. pre) I only have one patient 1, one patient 2, one patient 3... etc.. Is this the problem??
I also refer the DESeq manual example for pasilla Dataset.
> pasillaDesign
condition libType
untreated1 untreated single-end
untreated2 untreated single-end
untreated3 untreated paired-end
untreated4 untreated paired-end
treated1 treated single-end
treated2 treated paired-end
treated3 treated paired-end
In this design, they could sucessfully run. But in same way(like my model), for fixed single-end, they also only have one treated sample..
> cdsFull = newCountDataSet( pasillaCountTable, pasillaDesign )
> cdsFull = estimateDispersions( cdsFull )
Maybe, I am missing something... please help me..
I have a question regarding to the replicates for DESeq.
Here is my data design..
> head(data)
KYM_pre KYM_post GDY_pre GDY_post HKS_pre HKS_post KWK_pre KWK_post PYM_pre PYM_post SHM_pre SHM_post
A1BG 13 24 25 10 47 22 19 12 27 45 50 61
A2M 3 4 4 8 37 8 19 6 4 3 7 11
A2M-AS1 7 1 1 1 10 1 7 7 0 0 0 8
......
> meta
patient drug
KYM_pre 1 pre
KYM_post 1 post
GDY_pre 2 pre
GDY_post 2 post
HKS_pre 3 pre
HKS_post 3 post
KWK_pre 4 pre
KWK_post 4 post
PYM_pre 5 pre
PYM_post 5 post
SHM_pre 6 pre
SHM_post 6 post
=====================================
My goal is to find DEG responsing to the drug..
I tried two strategy neither of them worked..
1) Frankly, I don't have replicates.. so..
d_rare <- newCountDataSet(data, meta)
d_rare <- estimateSizeFactors(d_rare)
d_rare <- estimateDispersions(d_rare, method="blind", sharingMode="fit-only")
dh_fit1 = fitNbinomGLMs(d_rare, count ~ patient + drug)
dh_fit0 = fitNbinomGLMs(d_rare, count ~ patient)
Unfortunately, this model return NO DEG based on adjusted pvalue.
2) I assumes 6 patients having same disease are biological replicates for more solid estimation for dispersion
d_rare <- newCountDataSet(data, meta)
d_rare <- estimateSizeFactors(d_rare)
d_rare <- estimateDispersions(d_rare) # which means using default option (maximum likelihood estimation)
Again, in this second trial I got the following error
"None of your conditions is replicated. Use method='blind' to estimate across conditions, or 'pooled-CR', if you have crossed factors "
Does anyone have an idea? I have replications for patient(e.g. 1 1 2 2 3 3 44 ...etc) and drug(pre/post) too. why I am getting this message?? My suspection is that when I consider the cross design.. for example, for fixed factor(e.g. pre) I only have one patient 1, one patient 2, one patient 3... etc.. Is this the problem??
I also refer the DESeq manual example for pasilla Dataset.
> pasillaDesign
condition libType
untreated1 untreated single-end
untreated2 untreated single-end
untreated3 untreated paired-end
untreated4 untreated paired-end
treated1 treated single-end
treated2 treated paired-end
treated3 treated paired-end
In this design, they could sucessfully run. But in same way(like my model), for fixed single-end, they also only have one treated sample..
> cdsFull = newCountDataSet( pasillaCountTable, pasillaDesign )
> cdsFull = estimateDispersions( cdsFull )
Maybe, I am missing something... please help me..
Comment