Now, consider one dataset in detail. The dataset is an example for which the planted motif has a higher log likelihood ratio score than MEME's prediction, yet we argue that log likelihood ratio still doesn't work well as an objective function in this case.