Two of the methods we compare against, MEME and Gibbs Motif Sampler, are stochastic-search based algorithms. We run them requiring one motif instance per sequence and limiting the search to the primary sequence strand only, while leaving other parameters at their defaults. Gibbs Motif Sampler is run with 100 random restarts to allow for sufficient sampling of the search space, and MEME is allowed to execute its own algorithm to search the dataset for good starting points for EM. Note that Gibbs Motif Sampler failed to execute on three largest datasets, crp, ihf and purr when run on our local linux machines; these datasets were submitted through the web server.