Michael | 7 Oct 18:17 2012
Picon

Re: DESeq diagnostics

Dear Wolfgang,

Thanks for positive reply.
I am sceptic about the trend of MA plot, namely a slope to the left.

Normalization to many spike-ins is still tricky, and from my experience it
was really hard to simply get one size factor.

BTW, is it better to estimate dispersion in paires (treatment1 / control;
treatment2 / control, ...) or rather for a whole dataset in case of 1
control / treatment 1-6  (each sample in 2 biol rep).

Best,
Micheal

On Sun, Oct 7, 2012 at 3:31 PM, Wolfgang Huber <whuber@...> wrote:

> Dear Michael
>
> the plots look fine, as far as it is possible to tell. What exactly are
> you worried about?
>
> Normalising to the majority, as DESeq does, works well for experiments
> where the expression changes are sparse, or symmetric. If these conditions
> are not met, spike-ins offer an alternative, but they can be fiddly
> experimentally, and if there are only few, accuracy can be problematic.
> These problems are well known, there is no simple cure, you just have to
> deal with them.
>
>         Best wishes
>         Wolfgang
>
>
> Il giorno Oct 7, 2012, alle ore 1:28 AM, Michael <mllmmllmmllmm@...>
> ha scritto:
>
> > Hi,
> >
> >
> > I'll try to ask once again :)
> >
> >
> > My experiment is RNAseq of cells depletions from RNA decay factors, so
> one
> > should expect to see more upregulations. The libraries are ribodepleted,
> > paired end and stranded. After mapping with tophat I use HTSeq/DESeq
> combo
> > to discover DE genes (among tophat genes.gtf, rRNA not included)
> >
> >
> > I have problem with MA plots which are skewed (example attached), there
> is
> > a clear slope suggesting that more upregulation of genes of lower
> > expression. What should think about this?
> >
> > Diagnostic scatter plots of log ratio also look weird (second one is
> match
> > MA plot); PCA is ok, heat maps too.
> >
> > I also tried to compare DESeq normalization with normalization to
> spike-ins
> > present in the libraries, but the size factors assigned by DESeq seems
> > much more accurate; although it's unclear why ?
> >
> >
> > https://www.dropbox.com/s/0oykefjy1fvtq1i/1.pdf (MA plot)
> >
> > https://www.dropbox.com/s/r95oydftyeoz9mu/scatter.pdf (scatter plots,
> > second it for the MA plot)
> >
> >
> >
> > Cheers,
> > Michael
> >
> >       [[alternative HTML version deleted]]
> >
> > _______________________________________________
> > Bioconductor mailing list
> > Bioconductor@...
> > https://stat.ethz.ch/mailman/listinfo/bioconductor
> > Search the archives:
> http://news.gmane.org/gmane.science.biology.informatics.conductor
>
>

	[[alternative HTML version deleted]]

_______________________________________________
Bioconductor mailing list
Bioconductor@...
https://stat.ethz.ch/mailman/listinfo/bioconductor
Search the archives: http://news.gmane.org/gmane.science.biology.informatics.conductor


Gmane