3
$\begingroup$

I am using the limma package in R to do some analysis on a count data matrix. I use the voom function and that normally creates a plot with the mean variance trend line in it. Now I created also a mean variance trend plot, but it doesn't look normal.

Has anyone seen something like this before?

enter image description here

$\endgroup$
1
  • 1
    $\begingroup$ It shouldn't look like a normal because count data is not likely to have a normal shape. Poisson and negative binomial are common shapes and can be highly skewed. The counts could look approximately normal if they are sums of several iid count variables. $\endgroup$ Commented Jun 29, 2012 at 14:35

1 Answer 1

1
$\begingroup$

Usually you get a close enough distribution, without the wolverine claws.

In your case, transcripts within a particular count range possess a linear correlation with the square root of the standard deviation. It's definitely a technical problem, but your data may still be good. Check, first of all if it comes form a particular sample of your dataset (or group of samples). Then, check if the transcripts measured in that range are incorrectly measured (e.g. if you have a similar experiment, even in GEO, plot the absolute expression ranks against each other).

enter image description here

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.