Figure 4.

Varying the relative sizes of sets A and B. Our previous tests used |A| = 100. Now |A| is varied and the AUC is again calculated, averaged over 5 experiments. The maximum AUC for each test is shown by the black cross. The expected degree of sets A and B are equal in all cases. The only case where the algorithm produces a deterioration of the expression based ranking is where A = 500, which is half of the total network size. In all other cases, the algorithm increases the AUC compared to that achieved by pure expression ranking. The highest AUC is achieved in the case where |A| = 100.