Skip to main content

Advertisement

Figure 1 | BMC Microbiology

Figure 1

From: Predicting transcriptional regulatory interactions with artificial neural networks applied to E. coli multidrug resistance efflux pumps

Figure 1

Steps on designing the artificial neural networks as motif predictors. (A) Feed-forward and bi-fans instances were extracted from the E. coli regulatory network. Square nodes correspond to TFs, and circles to their targets that are either operons or isolated genes. (B) Generic examples of true FF/BF motifs and their counterparts. Non-motif samples were generated by modifying one or more targets of the real motif example, as exemplified in the highlighted orange nodes. (C) Procedures for assembling the feature vectors. Here, there is an example of how the BF motifs 1, 2, 3, 4 and BF motifs 1, 2, 5, 6 (illustrated in (B)) are encoded as vectors of correlations. These vectors store the correlations among transcript profiles of motif elements, for all possible pairwise combinations. The k(x, y), s(x, y) and p(x, y), are the Kendall, Spearman and Pearson correlation between x and y, respectively. Also, pc(x, y, z) and pc(x, y, z, t) correspond to the 1st and 2nd order Pearson partial correlation. Therefore, k(1, 2) is the Kendall correlation between the expression profile of TF1 and TF2, k(1,3) is the correlation between TF 1 and its target 3 (an operon or a gene). (D) Learning dataset and the neural network topology used in the study.

Back to article page