Re: DREME, DEME and discriminative MEME with -psp

[James Johnson]

I can't tell you about DEME as I've never used it but MEME and DREME work quite differently.

DREME is easiest to describe because it always has a negative set (it shuffles the positive set to make one when a set isn't given) and it just counts things. To find a motif it:

1. Determines all sub-sequences (aka words) that have a length between the minimum and maximum in the positive set
   
2. Counts how many sequences have each of those words (both the positive and negative sets are counted independently).
   
3. Applies the Fisher exact test to the counts of sequences that contain words in the positive set out of the total sequence count in the positive set and the counts of sequences that contain words in the negative set out of the total sequence count in the negative set.
4. Then it takes the most significant words and tries combining them into a IUPAC regular expression to see if it improves the significance
5. Finally to make the pwm it takes the IUPAC regular expression and finds all sites in the positive set that match it, these are combined to make the motif.

MEME doesn't take both sets so to enable the discriminative mode, instead it reads in prior probabilities calculated by psp-gen which basically bias a normal MEME search away from areas that are similar to the negative set. The advantage over DREME is that it can handle more complex motifs.

[Tonatiuh Pena Centeno]

Thanks a lot, your explanation was very helpful!!!


Tonatiuh