Hi, We are routinely using the DESeq package to identify miRNA differentially expressed in small RNA sequencing datasets from 2 biological conditions (A and B). Recently, we got interested in miRNA species that do *not* change. In a quick approach, we reasoned that miRNAs that do not change are those that are not significantly differentially expressed (Yes,we are big brain). But then I realized that because the H0 hypothesis upon DESeq usage is qi(A)=qi(B) it is not really possible to attribute a p-value to a "non-differential" expression I suspect (I am not a statistician) that to do this we should revert the H0 hypothesis in the DESeq procedure, something like qi(A) != qi(B). Does it make sense, and if so, is there any possibility to do this with DESeq ? Thanks for the help Christophe Antoniewski Drosophila Genetics and Epigenetics Institut Pasteur 25 rue du Dr Roux 75724 Paris cedex 15 Lab Tel: 33 1 44 38 93 35 Cell phone: 33 6 68 60 51 50 Fax: 33 1 40 61 36 27 Lab Web site: http://drosophile.org Post-Doc Position Available [[alternative HTML version deleted]]

Dear Christophe if you think about a set of possible worlds that are indexed by the parameter epsilon, and for which qi(A)=qi(B)+epsilon is true, then it would be difficult to construct a test that keeps its type I error (probability of false rejection) for all (nulls) epsilon!=0 but has any power for the (alternative) epsilon=0. Absence of evidence is obviously not the same as evidence of absence, but I think using the failure to reject the null hypothesis qi(A)=qi(B) as an absence of evidence for differential expression -as you suggest- is one of the better things you can do here. Wolfgang Il Apr/14/11 1:08 PM, Christophe Antoniewski ha scritto: > Hi, > > We are routinely using the DESeq package to identify miRNA differentially expressed in small RNA sequencing datasets from 2 biological conditions (A and B). > Recently, we got interested in miRNA species that do *not* change. > > In a quick approach, we reasoned that miRNAs that do not change are those that are not significantly differentially expressed (Yes,we are big brain). > But then I realized that because the H0 hypothesis upon DESeq usage is qi(A)=qi(B) it is not really possible to attribute a p-value to a "non-differential" expression > I suspect (I am not a statistician) that to do this we should revert the H0 hypothesis in the DESeq procedure, something like qi(A) != qi(B). > > Does it make sense, and if so, is there any possibility to do this with DESeq ? > > Thanks for the help > > > Christophe Antoniewski > > Drosophila Genetics and Epigenetics > Institut Pasteur > 25 rue du Dr Roux > 75724 Paris cedex 15 > Lab Tel: 33 1 44 38 93 35 > Cell phone: 33 6 68 60 51 50 > Fax: 33 1 40 61 36 27 > Lab Web site: http://drosophile.org > Post-Doc Position Available > > > > > > [[alternative HTML version deleted]] > > _______________________________________________ > Bioconductor mailing list > Bioconductor at r-project.org > https://stat.ethz.ch/mailman/listinfo/bioconductor > Search the archives: http://news.gmane.org/gmane.science.biology.informatics.conductor -- Wolfgang Huber EMBL http://www.embl.de/research/units/genome_biology/huber

Dear Christophe, With a classical statistical approach, you are correct that there is a difficulty in attributing p-values to non-differential expression. If we adopt a Bayesian approach, however, we can calculate the posterior likelihood of non-differential expression; you may therefore want to take a look at the 'baySeq' package, which can make such inferences. Best regards, Tom Hardcastle -- Dr. Thomas J. Hardcastle Department of Plant Sciences University of Cambridge Downing Street Cambridge, CB2 3EA United Kingdom > > ---------------------------------------------------------------------- > > Message: 1 > Date: Thu, 14 Apr 2011 13:08:47 +0200 > From: Christophe Antoniewski <christophe.antoniewski at="" pasteur.fr=""> > To: bioconductor at r-project.org > Subject: [BioC] DESeq - non-differential expression > Message-ID: <ff206021-ebcd-46cd-909a-b9a45416e7b4 at="" pasteur.fr=""> > Content-Type: text/plain > > Hi, > > We are routinely using the DESeq package to identify miRNA differentially expressed in small RNA sequencing datasets from 2 biological conditions (A and B). > Recently, we got interested in miRNA species that do *not* change. > > In a quick approach, we reasoned that miRNAs that do not change are those that are not significantly differentially expressed (Yes,we are big brain). > But then I realized that because the H0 hypothesis upon DESeq usage is qi(A)=qi(B) it is not really possible to attribute a p-value to a "non-differential" expression > I suspect (I am not a statistician) that to do this we should revert the H0 hypothesis in the DESeq procedure, something like qi(A) != qi(B). > > Does it make sense, and if so, is there any possibility to do this with DESeq ? > > Thanks for the help > > > Christophe Antoniewski > > Drosophila Genetics and Epigenetics > Institut Pasteur > 25 rue du Dr Roux > 75724 Paris cedex 15 > Lab Tel: 33 1 44 38 93 35 > Cell phone: 33 6 68 60 51 50 > Fax: 33 1 40 61 36 27 > Lab Web site: http://drosophile.org > Post-Doc Position Available > >