Background model for FIMO command line version

[atla goutham]

I am trying to run FIMO command line version to scan motifs against the entire hg19 genome. I have few queries about the background model.

1. The default background model is to use zero order NR sequence database. Does it means FIMO connects to the network to fetch these values ?

2. One of the previous posts suggests that FIMO can be used only with zero order background file. But I found a paper that used a fifth order file. http://www.nature.com/ng/journal/vaop/ncurrent/full/ng.3432.html

3. What would be the good background model if I am scanning for motifs for enhancer regions ?

4. To what extent the background model effects the results ? For example using the default background model vs a custom background model ?

Thanks and Regards,
 

[CharlesEGrant]

1. No, the NR background model is built into the source code. There isn't any code in FIMO that accesses the network.

2. FIMO and most of the rest of the MEME Suite tools only use a 0th order background model. MEME is the only tool in the MEME Suite that can use higher order models.

3. Ideally the background model should be derived from sequences that are biologically similar to the sequences you are scanning, but that don't contain instances of the motif you are scanning for. If you are using FIMO to scan selected enhancer regions you could try generating a background from unrelated enhancer regions for the same organism. If you are scanning all of hg19, then you should probably generate the background from all of hg19.

4. Providing an appropriate background model is the single most important step you can take to improve your FIMO results. The FIMO match score is the log-odds ratio of the probability of observing a sequence given the motif PWM to the probability of observing the same sequence given the background model.

[Hai Li]

Can I use the shuffle sequence of the target sequence as background?

[CharlesEGrant]

I'm not sure I understand your question. The 0th order background model is just the frequencies of the symbols. Shuffling a sequence doesn't change the symbol counts, so you'd get the same background model whether you generate it from the target sequence or a shuffled version of the target sequence. This page on the background model may be helpful.