Tools For Chipseq Scale Motif Finding?

Question

[Cassj]

Does anyone have any recommendations for motif-finding algorithms that can use ChIP-seq scale data? Approaches that can use all of the data, ideally incorporating ranking information, rather than just running MEME or similar on the top 50 ChIPseq peaks.

I'm also looking for methods that can handle motifs with substructure - for example I'm looking at some TFBS data and the binding site has left and right half sites which can appear on their own, or together in different orientations and with different spacers between them. Any suggestions on strategies for de novo identification of such patterns?

TIA, Cass.

Answer 1

[Brad Chapman]

There was a message on the Bioconductor mailing list last month describing a workflow for doing this in R:

Made up of 3 tools, 2 of which are in Bioconductor devel:

• PICS (ChIP-seq): I don't think this is in Bioconductor yet
• rGADEM (motif discovery)
• MotIV (motif validation)

Answer 2

[Mikael Huss]

I use CisFinder, which comfortable handles ChIP-seq scale data. It has a (pretty user-friendly) web interface, although I often use the command-line version as well.

[Mikael Huss]

I've also seen a program called HMS (http://www.sph.umich.edu/csg/qin/HMS/) meant for ChIP-seq motif finding, but I haven't tried it yet. There is at least one more out there.

[Cassj]

Awesome. Thanks Mikael :)

Answer 3

[Rob]

If you require ab-initio motif discovery, then perhaps DRIM may be of use. From memory, I don't think it's quite as sophisticated as what you're ideally after (left and right halves and so forth), but it does use ranking data. Possibly GLAM2 would be the most appropriate tool (if you need to represent a gap between two binding domains). GLAM2 does not take rank into account, though.

If you decide to use MEME on ChIP-seq data, you're best off running it in OOPS mode (according to our testing).

If you're able to scan the ChIP-seq peak regions with a library such as UniProbe or Transfac, then you have many more tools available that use rank information such as PASTAA or AME.

Disclaimer: I recently (on April 1st, what does that say?!) published AME in this paper.

It focuses on rank-based motif enrichment analysis methods (i.e. identifying statistically enriched motifs from a library of motifs). It compares several different MEA methods. Perhaps it (and the papers/tools it cites) will be of some use ;).

Answer 4

[Darked89]

These pages are a bit dated, but still should contain something usable:

• http://openwetware.org/wiki/Wikiomics:Sequence_motifs (see the top points)
• http://openwetware.org/wiki/Wikiomics:ChIP-chip (older)

In short, you may not be better by feeding probably any motif finding programs with hundreds of sequences. Going for sequences conserved between species is a low hanging fruit strategy, but should get you something to test fairly quickly.

Answer 5

[Jacques Van Helden]

Try the tool peak-motifs of the suite Regulatory Sequence Analysis Tools. http://rsat.ulb.ac.be/rsat/

For description, see

Thomas-Chollier M, Herrmann C, Defrance M, Sand O, Thieffry D, van Helden J. RSAT peak-motifs: motif analysis in full-size ChIP-seq datasets. Nucleic Acids Res. 2011 Dec 8.

Feeb-back and suggestions welcome

Jacques

Answer 6

[Valentina]

There is this algorithm that can deal with thousands of peaks:

ChIPMunk: http://www.ncbi.nlm.nih.gov/pubmed/20736340