The true positive rate vs. false discovery rate of two classifiers using both the RBF and TPPK kernels.

<p>The false discovery rate (FDR) was estimated assuming a negative-to-positive ratio of 600∶1. The black line indicates the performance of the classifier using all features, while the red line indicates the performance of the classifier without using the features based on GO term annotations. The <i>x</i>-axis begins at an FDR of 3% because very small FDRs cannot be estimated accurately for this dataset.