>program name
meme 3.0

I experimented with several parameters.  The main one is the motif occurrence model.  
The most flexible, TCM, seemed to give the best results in general for these data sets.  
I also tried varying the maximum motif width (16 or 24) and the order of the background 
model (1, 3 or 6).  In the end, I went with model=TCM, maxw=24 and order=6 on all data 
sets, but the results were qualitatively very similar using maxw=16 or order=1,3.

For each data set, I asked MEME to find three motifs, and then I went through and 
selected by eye the motif that I personally found most convincing.