Hello,

I have data of methylation percentages of tiles (from rrbs bisulfite experiment).

Is it possible to use the SVA package for batch effect correction? If so, what should the input be - the actual percentages?

Thanks

Hi Ray

I have heard of folks using SVA on this type of data, but they usually do it on something like the logit transformed percentage scale. I haven't actually tried it myself though so I'm not totally sure how it will work. The assumptions behind the sva algorithm require an approximately symmetric distribution that isn't bounded - percentages may not work great particularly for very low levels or high levels. 

Best

Jeff