Hi Mike, In fact have timecourse data (I attach an example here and hope it goes through). However, the confidence intervals shown in the figure are computed assuming a normal distribution. I am speaking only about normalized data, no ratio or other comparisons between two conditions. For the measurement I have replicates, I would like to plot them as the mean ? 95% confidence intervals. As I understand, but I might be wrong, the upper and lower confidence intervals shouldn't be symmetric as the data follow a negative binomial distribution (and if I am wrong here, my question makes no sense). Basically, I was wondering if it was possible to use the dispersion estimates calculated by DESeq to calculate the lower and upper confidence intervals and raw (normalized) read counts. I hope what I intend to do is now more clear. Yvan On Wed, Apr 10, 2013 at 6:04 PM, Michael Love <michaelisaiahlove at="" gmail.com="">wrote: > hi Yvan, > > To be a little more specific, are you looking for confidence intervals > for the log fold changes? or for the dispersion estimates? > > Mike > > > On Wed, Apr 10, 2013 at 5:04 PM, Yvan Wenger <yvan.wenger at="" unige.ch=""> wrote: > >> Hello everybody, >> >> It might be a very naive question but: >> >> As I can see, the estimateDispersions function can be used to used to >> retrieve the dispersions used by DESeq. From that data, is there a way to >> calculate upper and lower 95% confidence intervals? >> >> Best, >> >> Yvan >> >> [[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 >> > > -------------- next part -------------- A non-text attachment was scrubbed... Name: Locus_010420_Transcript_5.pdf Type: application/pdf Size: 5365 bytes Desc: not available URL: <https: stat.ethz.ch="" pipermail="" bioconductor="" attachments="" 20130410="" 8df7807a="" attachment.pdf="">

hi Yvan, This is not currently implemented in DESeq or DESeq2, but it is theoretically possible to get standard errors for fitted means for GLMs. For example, using glm.nb from the MASS package and then predict.glm with the argument se.fit = TRUE and providing the dispersion. As a side note, I might prefer a visualization of the normalized counts themselves, using dotplot or stripplots. Especially if there are few samples, it's good to have a sense of the 'raw' data (normalized only for size factors). Mike On Wed, Apr 10, 2013 at 6:54 PM, Yvan Wenger <yvan.wenger@unige.ch> wrote: > Hi Mike, > > In fact have timecourse data (I attach an example here and hope it goes > through). However, the confidence intervals shown in the figure are > computed assuming a normal distribution. > > I am speaking only about normalized data, no ratio or other comparisons > between two conditions. > > For the measurement I have replicates, I would like to plot them as the > mean ± 95% confidence intervals. As I understand, but I might be wrong, the > upper and lower confidence intervals shouldn't be symmetric as the data > follow a negative binomial distribution (and if I am wrong here, my > question makes no sense). Basically, I was wondering if it was possible to > use the dispersion estimates calculated by DESeq to calculate the lower and > upper confidence intervals and raw (normalized) read counts. > > I hope what I intend to do is now more clear. > > Yvan > > > > On Wed, Apr 10, 2013 at 6:04 PM, Michael Love <michaelisaiahlove@gmail.com> > wrote: > >> hi Yvan, >> >> To be a little more specific, are you looking for confidence intervals >> for the log fold changes? or for the dispersion estimates? >> >> Mike >> >> >> On Wed, Apr 10, 2013 at 5:04 PM, Yvan Wenger <yvan.wenger@unige.ch>wrote: >> >>> Hello everybody, >>> >>> It might be a very naive question but: >>> >>> As I can see, the estimateDispersions function can be used to used to >>> retrieve the dispersions used by DESeq. From that data, is there a way to >>> calculate upper and lower 95% confidence intervals? >>> >>> Best, >>> >>> Yvan >>> >>> [[alternative HTML version deleted]] >>> >>> _______________________________________________ >>> Bioconductor mailing list >>> Bioconductor@r-project.org >>> https://stat.ethz.ch/mailman/listinfo/bioconductor >>> Search the archives: >>> http://news.gmane.org/gmane.science.biology.informatics.conductor >>> >> >> > [[alternative HTML version deleted]]