This is a test version of Biostars. For the public version, visit https://www.biostars.org.
linear regression on EPICmethylation data

Hi all, I have a 2*2 study design where each subject got measured pre-exposed and post-exposed for low-exposure and high-exposure group. so it is 10 patients x 2 groups x 2 (pre & post)= 40 samples of methylation data. and i want to run linear model to see the differential methylation associated with exposure levels (numeric values). I saw a same statistical model from section 3.5 of the edgeR vignettes: comparisons both between and within subjects.

My question is can this statistical model apply to methylation data as well, although its designed for RNA-seq? Or is there any suggested R package to model my study design? Thank you so much in advance, Catherine

methylation r edger
• 2,148 views • link
updated by theobroma22 theobroma22 • written by catherine catherine

1 answer

Yes, a linear model is a linear model and can be used for any type of data like microarray, RNA-seq, methylation or other. The main issue is how to normalize your methylation data before putting it into a linear model. I'm not too keen on any bias or effects or the characteristics / properties of methylation data, so do you know if you can just mean center and scale it to the log scale as a normalization process if there isn't a different normalization method available?

Alternatively, and as a mere suggestion perhaps you could draw up your own null and full models, then compare them using ANOVA. Something in R would be like

null=lm(data ~ pre+post)
full=lm(data ~ pre+post+exposure)
aov(null, full)

This will tell you the effects of exposure.

• 0 views • link
written by theobroma22 theobroma22

Log in to answer this question.

More posts like this