[BioC] edgeR: plotBCV, gof() and plotMDS, for outlier detection

[Ryan]

Ok, since the upper band disappears when you exclude the outlier 
sample, I think that means that all the genes in the upper band of the 
original BCV plot are the specific genes that are responsible for the 
sample being an outlier, and the rest of the genes (the lower band) are 
behaving normally even in the outlier sample. You might try looking at 
the behavior of the 100 highest-dispersion genes in the outlier sample 
vs the other samples, to see if there is a consistent pattern (e.g. 
they are all way overrepresented in the outlier sample). Again, though, 
I'm not sure what you can do to fix the problem other than discarding 
the sample entirely.

On Mon Mar  3 00:11:26 2014, Sindre Lee wrote:
> On 2014-03-03 09:01, Ryan wrote:
>> That BCV plot looks bizarre, with the two bands. What does the BCV
>> plot look like if you exclude the outlier sample? Does the
>> double-banding go away? Does the prior.df change?
>>
>> Regardless, I think it's pretty clear from the MDS that this sample
>> is an outlier and should probably be excluded from your analysis.
>>
>> On Sun Mar  2 12:00:50 2014, Sindre Lee wrote:
>>> Hello!
>>> I have been struggling with one of the skeletal muscle biopsies in my
>>> study. The RNA quality is very good and looking at tissue specific
>>> gene expression they are all there, although some with very different
>>> values for some genes compared the other biopsies.
>>>
>>> Please see the attached plots using edgeR. The gof() calculation gave
>>> 0 "TRUE" in the $outlier table.
>>>
>>>
>>> Can anyone shed some light on this? Thank you very much.
>>>
>>>
>>> _______________________________________________
>>> 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
>
> Thank you!
>
> Yes, it goes away after removing the outlier (see attachment).
>