Figure 6
Performance evaluation of DynaSeq for individual TFs and in comparison to DNAshape and GC content based models for a 7-nt window. (a) Each TF is represented by a single AUC, which is the highest value from the 401 AUC values computed at each of the +/−200 nt positions from the motif start position of that TF. (b) Correlation between the best AUC values by DynaSeq and those by DNAshape. (c) A single cumulative performance level is obtained by averaging AUCs of all TFs at each of the 401 positions relative to their motif start site and shows how the perfomance levels vary when DNAshape and DynaSeq features from these positios are used in predictive models. However, such changes are not observed when GC content used. (d) AUC values plotted as a function of distance from motif. The values were calculated using data of Fig. 6(c). The plot gives a directionless estimate of predictability of TFBs from non-motif positions.
