Is there an informative subset of 450k methylation probes?

Question

[rmccloskey]

I'm working on an analysis which includes 450k methylation data. There are so many probes that analysing the whole data set is becoming a problem in terms of time and memory. I'm sure that nearby methylation sites are highly correlated, so is there some kind of informative subset of the whole probset I could use, to reduce computational costs without losing too much information? I'm aware that it's possible to do this myself using clustering or something, but I was hoping it had been done already.

Answer 1

[Charles Warden]

While the correlation of nearby CpG sites is an assumption made in the probe design, I think it is best to take advantage of as much information as possible.

I (and others) have done some work on trying to define differentially methylated regions from 450k data. I have some templates for analysis for a couple programs here:

• http://www.nature.com/protocolexchange/protocols/2965#/introduction
• http://sourceforge.net/projects/cohcap/files/Protocol_Exchange_Example.zip/download

How many samples do you need to analyze? For small cell line datasets, I think the above tools should be OK for most desktops (but I agree that large patient cohorts may need to be run on a more powerful Linux cluster).

[rmccloskey]

I have 389 samples. I agree that it's best to use as much information as possible, but I'm already running on a cluster and am still having memory issues.