Question: Illumina 450K analysis with a probe selection
gravatar for IOM
3.4 years ago by
United Kingdom, Birmingham, NHS
IOM0 wrote:


Thanks for reading, I would like to ask you for your opinion. I would like perform a methylation analysis comparing two groups of samples with a determined number of probes/genes, not using all the probes available in the array (this gene list comes from a literature search). My questions are:

  • Should probe selection be done before the contrast? Reduce the number of probes to a few hundreds/thousands before any comparison.
  • Should probe selection be done after the contrast? Use all the probes and do a heatmap, for example, only with selected probes.
  • What statistical consequences any of this selections would have?
  • Does anybody know how to perform the probe selection in a workflow using minfi/limma, for example? I am asking this because the annotation of the probes is done after performing the contrasts.

Any help will be much appreciated. Thanks


methylation illumina 450k • 1.3k views
ADD COMMENTlink modified 3.4 years ago by andrew.j.skelton736.0k • written 3.4 years ago by IOM0
gravatar for andrew.j.skelton73
3.4 years ago by
andrew.j.skelton736.0k wrote:

Assuming that you've applied a normalisation process with Minfi, and you have a resulting GenomicRatioSet class object, you can get the annotation using:

norm_data_f         <- norm_data_f %>% addSnpInfo
norm_data_f         <- norm_data_f %>% dropLociWithSnps(., snps = c("SBE","CpG","Probe"), maf  = 0.05)
annotation          <- norm_data_f %>% getAnnotation %>%

Which gives you a full set of annotation for your probes, prior to modelling with Limma. The GenomicRatioSet is also a subsettable S4 class, so you can use the annotation to provide sensible filtering. Minfi provides a nice helper function dropLociWithSnps for probe level filtering.

Generally speaking, I wouldn't cherry pick genes to test, I'd run the contrast across all probes passing your probe level filters. Pre-selecting genes to test means that you reduce the total number of tests, and the resulting FDR is more lenient.

ADD COMMENTlink written 3.4 years ago by andrew.j.skelton736.0k

Hi Andrew,

Thanks very much for your response. As you say, reducing the number is not appropriate and I would expect those genes to be differentially methylated using all probes if they truly were (without the reduction), is that correct? Thanks


ADD REPLYlink written 3.4 years ago by IOM0

That is correct, yes.

ADD REPLYlink written 3.4 years ago by andrew.j.skelton736.0k


ADD REPLYlink written 3.4 years ago by IOM0

If an answer is helpful it's appropriate to upvote it. If the answer resolved your question, you should mark it as accepted.

ADD REPLYlink written 3.4 years ago by WouterDeCoster44k
Please log in to add an answer.


Use of this site constitutes acceptance of our User Agreement and Privacy Policy.
Powered by Biostar version 2.3.0
Traffic: 1661 users visited in the last hour