Figure 4

Cross Validation with Receiver Operating Characteristic (ROC) curves reveals increased predictive power over weight matrices. 1000 iterations of 10-fold random-split cross validation reveal that the most predictive models utilize the correspondence structures. The combined-palindromic model is the best model since it contains roughly half as many parameters as the other three correspondence models. The positive set contains 160 experimentally validated p53 binding sites, and the negative set contains 40 bp random samples from the mononucleotide content of the training set. The true positive and false positive rates are calculated and plotted for all possible threshold values for each model. The predictive measure for comparing the curves is the AUC (Area Under the Curve). In all the PHMM models the insert-state emissions are fixed to the A, G, C, T nucleotide distribution of the training set. The best classifier uses the combined-palindromic training motif. (Position ã has the complement nucleotide emission distribution of a).