Skip to main content

Table 6 AUC of SVM on the testing set. The testing set is comprised of 50 functional categories classified into 10 groups according to how informative they are. A GO function is said to be more informative if fewer numbers of genes are associated with it. An SVM was trained for each of these categories using two sets of parameters. In this table, Group 1 consists of the most informative functions whereas Group 10 consists of the least informative functions. The "default" column reports to the prediction performance on the AUC scale using default control parameters in our SVM software Gist. The "tuned" column reports the prediction performance on the AUC scale using control parameters optimized on the corresponding the training set. The AUC values listed are averaged across the four C-V folds. See the substraction "SVM" Parameter Selection" in the "Method" section for further detail.

From: A factor analysis model for functional genomics

Group Default Tuned
1 0.804 0.793
2 0.683 0.638
3 0.803 0.806
4 0.853 0.862
5 0.864 0.863
6 0.751 0.740
7 0.788 0.805
8 0.807 0.796
9 0.748 0.773
10 0.739 0.766
Mean 0.784 0.784
(Stdev) (0.055) (0.064)