We also show sSn and nPC values while ignoring significance for each of the three other methods compared to LP/DEE in Figure 1, only displaying transcription factor datasets for which a difference in performance is observed. Each bar in the chart measures the difference in sSn (Figures 1(a)–1(c)) or nPC (Figures 1(d)–1(f)) between our method and one of the other methods. Using both the sSn and the nPC statistics, LP/DEE performs better than any of the three other approaches in identifying known binding sites. For example, for LP/DEE versus MEME, very large differences are observed for three transcription factors, with our method identifying narL, glpR, and ntrC motifs almost completely, and MEME entirely misidentifying them. Moreover, the LP/DEE method exhibits better performance on more transcription factor datasets than the other methods. For example, considering nPC, LP/DEE performs better than MEME on eleven datasets, and worse than it on six datasets (Figure 1(e)). Differences in performance with Gibbs Motif Sampler and Projection are smaller; for instance, the LP/DEE method exhibits better performance than Projection using the sSn statistic on six datasets versus worse than it on two datasets (Figure 1(c)). We note that if significance assessments are included and motifs with e-value greater than 1.0 are discarded (see Additional Figures 1(a) and 1(b)), then LP/DEE has better nPC than MEME on 16 datasets, and worse nPC on three datasets, suggesting that MEME's significance computation is unnecessarily conservative for our dataset; the same applies to Gibbs Motif Sampler as well.