Figure 2.

E. coli FruR motif sequence logos and occurrences.(a-e) Known and inferred E. coli FruR motif sequence logos. (a) The known E. coli FruR motif. The central part of the motif has a number of well-conserved positions; however, the outermost positions are very poorly conserved and may be incorrectly statistically regarded as background. A heuristic for determining the most likely width is required to be robust in statistically unclear situations such as this. (b) A motif discovery algorithm may become locked in a non-optimal local maximum of the likelihood function which corresponds to a shifted version of the true motif. (c) The most statistically significant model in a set of candidate models may only represent a portion of the true motif. (d) From the candidate set of computationally discovered models, MCOIN chooses the model at w^∗ + 1, which corresponds well with the true motif. (e) The E-values estimator chooses the model at w^∗ - 3, which corresponds less well with the true motif. (f) Known occurrences of the E. coli FruR motif.