Figure 3

Comparison of motif finders in their ability to identify true LexA binding sites. A dataset of constructed of 30 LexA binding sites and 19 random 1000-mers taken from around the E. coli genome was evaluated with 2 popular motif finders (MEME ZOOPS and Gibbs recursive sampler) and each of the 3 available MotifCatcher options (MEME ZOOPS MC, Single MAST MC, Iterative MEME/MAST MC). MotifCatcher results were selected with an FP Threshold value based on the lowest E-value reported. Results from all 3 MotifCatcher frameworks had better specificity (column 3) than the MEME ZOOPS and Gibbs recursive sampler results, and the iterative search (Iterative MEME/MAST MC) had the best F-measure of all 5 options. The motifs produced by the Gibbs recursive sampler results and all 3 MotifCatcher runs demonstrated a pronounced T-TATA upstream of the CAG (column 5) not seen in the MEME ZOOPS results. This difference represents a de-emphasis in palindromicity of the motif, suggesting a directionality of LexA protein binding in vivo. Note that the MotifCatcher runs produced motifs with E-values 20–27 orders of magnitude than the MEME ZOOPS result, which corresponded to an improvement in F-measure of about 0.25.