Questions about creating epigenetic priors for FIMO (psp)

[Amira Kramdi]

Hello,

I have a couple of questions about FIMO's --psp option :

Is there a recommended resolution for the input wiggle file used for create-priors ? Does it have an impact on FIMO's predictions ? I plan on using H3K27ac mark and perhaps ATAC-seq data in the future.

Is it necessary to start with raw densities or is it ok to work with read densities normalized by sequencing depth ?

Finally, in the paper it is reported that DNase I prior achieved the best improvement in TFBS identification compared to histone marks. This makes me wonder why histones marks priors alone do not perform this well. It is because the valleys in histone marks, which are indicative of a potential binding sites, do not contribute as much because of the drop in coverage ? In this case FIMO might focus on segments near the valleys rather than the valley itself.. Does it make sens ?

Thanks in advance for the clarifications,

Best,

Amira

[cegrant]

Is there a recommended resolution for the input wiggle file used for create-priors ? Does it have an impact on FIMO's predictions ? I plan on using H3K27ac mark and perhaps ATAC-seq data in the future.

The conceptual goal of FIMO is to identify the locations in a sequence where a transcription factor will bind. What FIMO actually accomplishes is to identify the positions in a sequence that closely match the position weight matrix for a transcription factor. Unfortunately this leads to lots of false positive and false negatives. A site can be an excellent match to a motif but not be a functional binding site and visa versa. We're just hoping that, in general, the better the match to the motif the more likely it is to be a functional site. If we have additional information about the biology of the site, that can help  us reduce the false negatives and false positives. This is where the priors come in: they identify sequences where transcription factor binding is actually likely to occur.

DNase I assays looks for the short sequences that have been protected from degradation by the mechanical action of having a protein molecule bound to the DNA, blocking cutting. The area protected is very limited and directly tied to the binding location of the protein. Histone marks are associated with more general phenomena like up or down regulation, and while they may include the binding site for a transcription factor, they will typically extend over a longer segment, so they don't help to narrow down the location as much. I'd guess that ATAC-seq would provide better results for the same reason: chromatin accessibility would be more specifically associated with the ability of a protein to bind the DNA than other epigenetic marks. This is just the guess of a programmer though, not a biologist.

For the same reason, higher the resolution the better. The higher the resolution the more weight will be given to good motif matches in those (short) regions with a high prior.