Motif Finding For Chip-Seq Data Using Peak Profiles?

Question

[2184687-1231-83-]

Are there any tools that do motif finding or motif discovery using the information contained in the peak profiles? That is, that they use the shape of the peaks as part of the information to decide where the motifs might be.

For example, a simplified case from a BAM file that has a peak with such shape:

                                 ....
                               ...   ...
                              ..       ..
                              .         ..
                              .          ..
                             ..           .
                           ...            ..
                    ........               .
                 ...                       .
                 .                         .
                .                          .
               ..                          .
              ..                           .
            ...                            ..
         ...          %             *        .
    ----------------------------------------------

In this example, the shape of the two peaks is taken into account when considering potential motifs, here the main one '*' with the highest intensity, but with another motif '%' also found to have a small summit next to it.

Is any of the existing programs explicitly using this sort of information? I would have guessed that it would help, specially when looking at co-occurrence of motifs in cis regulatory modules (CRM) from multiple chip-seq experiments.

[brentp]

nice ASCII art!

[Ying W]

maybe create two bed files with varying contingencies for peak so that one would only have * and the other would have % and *

Answer 1

[Mikael Huss]

There is one tool that does something like what you are asking for, but unfortunately I don't think it has been published by itself yet. The method is called TherMoS and is described to a certain extent in the supplementary material to this paper: http://www.nature.com/msb/journal/v6/n1/synopsis/msb2010109.html

TherMoS "employs nonlinear regression to fit the observed distribution of sequence tags in a ChIP-seq library, and produces an estimate of the position-specific energy matrix (PSEM), i.e., the matrix of binding free energy changes induced by all possible single mutations of the consensus-binding sequence."