Figure 4

Function prediction using regulatory motifs in T. brucei. This figure shows inositol phosphate metabolism (KEGG:tbr00562) as an example. (A) The performance of our naïve Bayesian network using different numbers of motifs for prediction of inositol phosphate metabolism genes. We used a two-fold cross-validation for assessing the prediction power, where half of the dataset was used for training and the other half for validation. Cross-validation was repeated 100 times for each number of motifs, and each time the AUC (area under the curve) of the ROC curve was measured as the prediction power. Standard deviation of AUC is shown by the error bars. (B) The ROC curve for prediction of inositol phosphate metabolism genes using all 36 predicted motifs. Standard deviation of sensitivity is shown by the grey shaded region. The diagonal line shows the performance that would be expected if our naïve Bayesian network was not able to predict inositol phosphate metabolism genes. This classifier has a very high specificity (~99%) at sensitivities of up to 20% for this pathway.