Figure 6:

We use repeated 10-fold CV for model training and testing. We repeat each CV 10 times, ensuring that fold membership changes for each run. We report the mean and standard deviation of eloquent class true positive rate (TPR), and eloquent class area under the curve (AUC). For each baseline, we report the FDR corrected p-value from the associated t-score between our MT-GNN and the baseline, as evaluated on the AUC metric. In addition, we report the specificity, F1 and t-scores for the main classification results shown in Tables 3 and 4.