Motif prediction in Chip-seq data

Question

[anu014]

Hello Biostars,

Which tool would you prefer for motif prediction in chip-seq data - HOMER, MEME or ChIPMunk? Is there any disadvantage of using a conventional tool for motif prediction of specific data such as ChIP-seq?

Thanks in advance :)

[venu]

What do you mean by "conventional tools" ? The tools you mentioned ? HOMER and MEME are completely fine (never used ChIPmunk). If you already have a set of TFs/PWMs you can use fimo from MEME, it is very fast (command line version).

If you have TF chip-seq data, it is okay to search for motif occurrences within peak regions. If you are trying to predict TF binding sites using Histone modification data (such as H3K27ac), you need to reduce the search space by restricting search to NFR regions.

[anu014]

Thanks for the reply Venu. Yeah, conventional tool such as HOMER as it doesn't have known motifs for bacteria n some other species, so I can't restrict the search just for bacteria.

[venu]

In such case, download motifs from other sources (check this) and use fimo.

Answer 1

[simon.vanheeringen]

All of them! :) This is why I wrote GimmeMotifs. It uses an ensemble of motif prediction tools and chooses the best non-redundant motif from the results. From benchmarking on ENCODE ChIP-seq data, Homer, MEME and BioProspector show the overall best performance. However, there is no single tool that performs best in all cases. For instance, if you have a very strong motif (present in almost all peaks), the best tool is different from when you have a much weaker motif. Until we come up with much better algorithms, I think using an ensemble is the way to go.