[BioC] edgeR and FDR values always equals 1

Simon Anders anders at embl.de
Fri Nov 12 19:39:13 CET 2010


Hi Anton

On 11/12/2010 06:33 PM, A Gossner wrote:
> Have plotted two samples as you suggested [ plot
> (d$counts[,2],d$counts[,3],log="xy")] and saved the file on the FTP server.
> ftp.ed.ac.uk
> Login using the username 'anonymous' and use your email address as
> password.
> get /edupload/logplot.eps as well as get /edupload/plot2.eps and get
> /edupload/plot2.eps

 > There does seem to be some correlation on the log-log plot.

Yes, there is some correlation but it's not great. If you have, say, 100 
counts in one sample, you seem to get anything from 0 to 3000 counts in 
the other. Hence, even if you get 5-fold changes between treatment and 
control, this will not be significant.

> Which are plot asinh(count) graphs you suggested but must confess not
> sure if plotted correctly, plot(d$counts[,2],asinh(d$counts[,2]))
> plot(d$counts[,3],asinh(d$counts[,3]))
> but they are plots 1 and 2 respectively.

This is not what I meant. It was rather something trivial: If you just 
want to compare correlation, then instead of

   plot (d$counts[,2],d$counts[,3],log="xy")

it can be nicer to use

   plot( asinh(d$counts[,2]), asinh(d$counts[,3]) )

because the zeroes don't get suppressed. (The area hyperbolic sine is 
linear for very low values and then becomes quickly indistuingishable 
from a log. hence this is a cheap way to get a log-log-like plot with 
zeroes. The axis tick marks are completely misleading though.)

   Simon



More information about the Bioconductor mailing list