Skip to content

MeDIP-seq data analysis – diffReps

February 19, 2014

MeDIP-seq (and this should be the same for methylCap-seq/MBD-seq which I’m told is better) is advertised as reasonable cost-effective method to find differentially methylated regions (DMRs). 40 million reads or 1/5th of a lane on the HiSeq is an adequate amount of sequencing depth. I was even told by the same guy as above that he found pretty much the same set of DMRs using methylCap-seq as he did methylC-seq.

About two years ago now, I did some meDIP-seq on some cancer cell lines. The data looked good. The one locus (KRT7 promoter/CpG island) we thought had a high chance of being a DMR between our two groups of cell lines was clearly differentially methylated. CpG islands that you would assume would be unmethylated (e.g. GAPDH) had no signal while loci that should be methylated showed high signal. But the problem was finding DMRs between our two classes of cell lines. I tried MEDIPS and MACS but they do not handle replicates natively and there was a lot of inter-condition variation, which just resulted in a lot of noise. I also saw that some minor changes in peak shapes could result in DMRs being called. Then I tried using HTseq-count to map reads to CpG islands, promoters, H3K4me1 peaks, etc. and then using DESeq to do the statistics. This worked a little better in that it did find a couple DMRs and it did return the KRT7 CpG island as a DMR and a couple others but it was a targeted approach to a small fraction of the genome.

After sitting on the data for longer then anyone should sit on data (about a year), a new ChIP-seq analysis package called diffReps was released in July of 2013 that handles replicates natively. Busy with my new job making TALEs and doing whole genome bisulfite sequencing, for months it was one of those things I kept on thinking I had to try. Finally, sometime towards the end of last year, I gave it a try and it work magnificently. It found a nice set of 2656 DMRs. Looking at them in the genome browser they were all nicely repeated across the our replicates and DMRs were preferentially found near our set of differentially expressed genes. So we finally got that piece of data we really needed and now the manuscript is just about finished.

So after doing meDIP-seq, my experience is you really need replicates. Our system had a lot of inherent noise but I think it’s a method that needs replicates regardless. I guess all experiments need replicates but some more then others. Also, it is definitely not a full genome-wide method. Areas with low CpG density do not get pulled down regardless of methylation status. I guess in theory you would eventually get coverage of low CpG areas if you sequenced deep enough but that would defeat the cost saving advantage of the method over whole genome bisulfite sequencing. And of course the resolution is not very good. You’re only going to see differentially methylated regions and not differentially methylated cytosines. But it gave us the data we needed. You can do a lot of samples without blowing the whole lab budget. And a of course a final thanks to the authors of diffReps for coming up and sharing a method that can handle the data.

ADAM22_ucsc

Advertisements

From → Posts

Leave a Comment

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: