Flowchart of the GLECLUBS algorithm for genome-wide de novo CRBS predictions.

The top 20 motifs/clusters predicted in B. subtilis. The logo is for the best motif identified by MEME in each cluster.

The top 20 motifs/clusters predicted in E. coli K12. The logo is for the best motif identified by MEME in each cluster.

The correspondence between the clusters and known motifs. (A) Distribution of the number of known motifs in which the sequences of a cluster are located. (B) Distribution of the number of clusters that contain the known binding sites of a motif.

Three distinct sub-motifs of the CRP binding sites correspond to three top-ranked clusters. The all known CRP binding sites (A), can be subdivided into a more information-rich canonical palindromic sub-motif (B), a T-rich motif (C) and an A-rich sub-motif (D). Our predicted eighth (E), 72-th (F) and 178-th (G) clusters, largely correspond to the CRP's canonical palindromic sub-motif (B), the T-rich motif (C) and the A-rich sub-motif (D), respectively.

Construction of the motif similarity graph. Multiple (k) motif-finding tools are used and each predicts multiple motifs in each inter-operonic sequence set I_o. All the predicted motifs form the set of input motifs. A node in the motif similarity graph represents a motif. Two nodes are connected if their corresponding motifs have a similarity score ≥β with the score being the weight on the edge (not shown for clarity). A solid line represents an edge between two nodes associated with two different I_o, and a dashed line represents an edge between two nodes associated with the same I_o.

Evaluation of the top-ranked clusters. (A) Cumulative recovery rate of the known binding sites in the input motifs by the top-ranked cluster, computed as the ratio of the number of cumulative known binding sites recovered in top-ranked clusters to the number of known binding sites in the set of input motifs. (B) Cumulative recovery rate of the known motifs that have binding sites in the input motifs by the top-ranked clusters, computed as the ratio of the number of cumulative known motifs recovered in top-ranked clusters to the number of known motifs in the input motifs. (C) The number of cumulative unique putative binding sites in the top ranked clusters as a function of the rank of clusters. (D) The number of predicted binding sites of E. coli K12 in the top 1000 clusters. (E). The number of known binding sites recovered by the top 1000 clusters in E. coli K12 by each cluster (dashed line), and by each 20 consecutively ranked cluster group (solid line). (F). The distribution of the number of putative binding sites of the top 400, 300 and 300 clusters in the inter-operonic regions of E. coli K12, B. subtilis and S. oneidensis, respectively, and the distribution of known binding sites in RegulonDB in the inter-operonic regions in E. coli K12.

The distributions motif similarity scores. (A) Comparison of the distributions of motif similarity scores computed by average Kullback–Leibler (AKL), average log-likelihood ratio (ALLR), Pearson correlation coefficient (PCC), 1 − P-value of Chi-square (pCS), sum of squared distance (SSD), asymptotic covariance (AC), and our metric. The alignment defined by our method (see ‘Materials and Methods’ section) between two motifs was used to compute the similarity scores by all the metrics. The scores computed by each metric were normalized to their maximal values. The solid lines are the distributions of the motif similarity scores among the input motifs in E. coli K12. The dashed lines are the distributions of the motif similarity scores among the sub-motifs of known binding sites in RegulonDB. (B) The overlapping area between the distribution curve of the similarity scores among the input motifs and that of the similarity scores among the sub-motifs of known motifs in RegulonDB. (C) Effects of the selection of reference genomes and the way of grouping inter-operonic sequences on the distribution of the similarity scores among the input motifs. The dashed lines are the distributions of the motif similarity scores among the sub-motifs of the known binding sites in RegulonDB or DBTBS. The vertical dotted line indicates the point of the similarity score cutoff β = 0.05 used for constructing the initial motif similarity graphs. (D) Distributions of the absolute values of Pearson correlation coefficient scores of the expression vectors of each pair of genes in each of the top 400 predicted clusters/regulons in E. coli K12 and that of randomly selected 400 gene groups with the corresponding numbers of genes in the top 400 predicted clusters/regulons.