Motif analysis of DMRs (Differentially Methylated Regions) by HOMER

Question

[inschigu]

I tried to use HORMER findMotifsGenome.pl) to identify TF binding sites on differentially methylated regions (DMRs), but I have no idea about the following three parameters.

Strand information There is not the strand information in DMRs. Is the BED file without the strand information acceptable for HOMER?

Size parameter (-size) Unlike ChIP-seq data, DMRs don't have a certain peak. In this case, is the "-size given" appropriate as "-size" parameter?

Sequence normalization options Since DMRs are basically CpG rich regions, is the “-cpg” option appropriate to normalize sequence?

Thank you

[pepinme]

I am also having the same question, as I am trying to develop a workflow from DMR multiple sequence alignment and clustering to de novo motif discovery and enrichment.

Answer 1

[Charles Warden]

You bring up a nice question about the -cpg option (in HOMER).

I there there are some common GC-rich motifs that HOMER tries to place less emphasis on. However, you specifically want to know about the CpG sites within the region (and sequence outside CpG motifs would not be as relevant).

If you are going to try and do some motif analysis, I think you would need to be able to find differential methylation across CpG-overlapping motifs at multiple genes.

So, you probably need to get site-centric transcription factor binding information. I think ENCODE would be a popular, but I can't precisely vouch for it's use in this application.

If you have that, I can think of at least two possibilities (methods-wise):

1) I believe methylSig performs such a test with ENCODE Motifs (section 6.3 in the linked PDF)

2) If you use an annotation-based strategy that doesn't take proximity into consideration, you could also do something similar (with your own set of annotations). Again, I don't have a success story about defining the "region" as CpG sites within motifs in different genes, but you can use the default COHCAP region analysis (without re-defining boundaries) to test this sort of thing (with the extra effort of defining your own "custom" annotation with platform="custom" in COHCAP.annotate())

[luisjavierleandrogarcia]

Hi, same questions here. Did any of you solve them? Thank you.

Answer 2

[TheEyeOfTheUniverse]

Hey!

About the strand if you use fasta input instead of bed it would solve you that problem. About the normalization, It's not a direct solution but you can use a background of random CpG sites. For the size parameter I don't see why you'll need it for motif enrichment.