DESeq for a small sets of sequences without replicates

[vebaev]

Thanks Simon for clearing this,
I'm not so into statistics, so I just used the methid with chi-square like they use it in several miRNA tools for example miRNAKey (http://bioinformatics.oxfordjournals...6/20/2615.long)

Vesko

[Simon Anders]

I do not know in your case, but in my case was miR analysis and I got no replicates, than I used chi-square test after normalization and p-values are more nicer. I also added a small number to each entry that way the ratio between low read counts will be not so significant (http://bib.oxfordjournals.org/content/10/5/490.abstract)

But for RNAseq maybe is different?

Of course you get more hits with a chi-squared test, because this means that you set the biological variance to be zero. Your results will not be reproducible, though.

I've explained this at length in this post: http://seqanswers.com/forums/showpos...04&postcount=2

In the terminology of this post, you tested for question (A) but should have tested for question (B).

Simon

[starlight]

Hi.
In my case it is also microRNA analysis. I will check this paper you sent.
Thanks.
Ana

[vebaev]

I do not know in your case, but in my case was miR analysis and I got no replicates, than I used chi-square test after normalization and p-values are more nicer. I also added a small number to each entry that way the ratio between low read counts will be not so significant (http://bib.oxfordjournals.org/content/10/5/490.abstract)

But for RNAseq maybe is different?

[starlight]

Hi vbaev.
Thanks for your answer. Yes, first all the p-values are quite high and also, those with pow one have a low read count.
Do you know whether it is due to the lack of replicates only or the combination of the lack of replicates and low size of the set of tags.

Thanks again,
Ana

[vebaev]

With no replicates it is quite difficult. My experience with no replicates was that I got lot of my entries with high p-value and it was useless.

[starlight]

Hi SEQanswers community.
I was trying to use the DESeq package to analyse differential expression between two RNASeq experiments, but these sets are rather small (around 130 points) and are very variable. On the top, I have no replicates for neither of two sets.

The results seem quite awkward, DESeq having chosen the points with very low number of reads as the ones with the lowest p-value.

I hence wondered whether DESeq is suited whatsoever for the use on the small datasets with no replicates.

Thanks in advance for any insight,
Ana