[BioC] Stange tailing in a Volcano-plot

Gordon Smyth smyth at wehi.edu.au
Thu Nov 6 23:14:37 MET 2003

At 07:54 PM 6/11/2003, Valtteri Wirta wrote:
>Dear Group,
>
>I have a question regarding the Volcano-plots which can be obtained after
>an analysis using the linear models in the limma package.
>
>In some cases I see a strange tailing phenomena in the Volcano-plot (x=M
>and y=eb\$lods). I think this is  a mathematical "artefact" and would
>appreciate if someone could explain this to me.
>This tailing looks like an inversed "normal distribution plot", and where
>I by inversed mean an up-side-down distribution plot with the lowest
>eb\$lods values at M=0.
>The image is available as a pdf file at
>http://biobase.biotech.kth.se/~valtteri/R/VolcanoPlot.pdf
>
>What I noticed is that the features in this tail have their s2.post values
>equal to 0.0663 in the object obtained using the ebayes function of the
>limma package. In the image provided I have coloured all features with
>s2.post=0.0663 in red.

These genes have no degrees of freedom to estimate a standard deviation,
presumably because 4 out of the 5 arrays have missing values for those
genes. So the posterior standard deviation for those genes is set to the
estimated prior standard deviation. So the t- and B-statistics for those
genes are a function of the M-value only. So you see the mathematical
pattern in the plot.

Gordon

>Some background on the analysis I did:
>R version: 1.8.0
>Limma version: 1.2.8
>The data is background subtracted, filtered and normalised data from 5
>slides with 30 000 features (no duplicates), stored in a numeric matrix, Ms
>fit <- lm.series(Ms, design=c(1,1,1,1,1))
>eb <- ebayes(fit)
>M <- fit\$coef   # to get the M values
>plot(M, eb\$lods)        # to give the volcano-plot
>
>I'd really appreciate if someone could explain strange (?) behaviour to me.
>
>Thanks!
>
>best regards,
>Valtteri
>
>
>
>
>
>
>Contact information:
>
>Valtteri Wirta
>Department of Biotechnology, KTH
>AlbaNova University Center
>S - 10691 Stockholm, Sweden
>