Measuring transcriptional noise

Question

[4galaxy77]

Hi, I have an RNAseq dataset for ~ 20k genes and 3 conditions, with 6/7 replicates within each conditions. I'm interested in looking at the amount of transcriptional noise, or variation in gene expression between castes. I've done some rough calculations, but I'm no statistician so I'm sure there are lots of errors in what I've been doing.

Does anyone know of any software that I can use which give me a more rigorous/stable method of calculating transcriptional noise? I've had a good search but can't seem to find any.

Thanks!

[Sean Davis]

What is the biological question? I assume that you want to measure the differences in "noise" between groups, not just "noise" across groups?

[4galaxy77]

Yeh sorry, I mean between groups. It's a side analysis of the main project, but essentially we have 3 different castes (i.e. 3 diff treatments) of within an insect species and it's been suggested that noise in transcription represents phenotypic plasticity - so we know that some of the castes show more plastic behaviours than others, so we want to see whether this is represented in an increase in transcriptional noise.

[pld]

Are these biological or technical replicates? Wouldn't you need both to know if the noise you're seeing is at the biological level and not imparted by the sequencing?

[4galaxy77]

biological replicates. yeh exactly, which is why im asking for a program which might do those kind of calculations :)

[Michele Busby]

Did you squash up whole bugs and pool them? You get some noise if the size or morphology of the bugs are variable that is hard to parse out from transcriptional noise. There is a paper on that... uh... somewhere... I could not find it with a quick google search but I believe it was in drosophila. But I would think about if that is too much of a confounder before you go too far down that road.

[4galaxy77]

No they were sequenced individually and then grouped into castes after they were sequenced. I've already removed the batch effects using a function in NOISeq. Plus they are eusocial hymenoptera so there's not gonna be much size variation between them.

Answer 1

[Chris Cole]

We have a paper looking at this:

http://bioinformatics.oxfordjournals.org/content/early/2015/07/22/bioinformatics.btv425

Technical replicates are essentially Poissonian and biological replicates are over-dispersed which can be best represented as either Negative Binomial or Log Normal distributions. Any 'bad' replicates will skew the noise dramatically.

Answer 2

[Sean Davis]

While this paper discusses methylation data, it may be of interest.