2.8. Binary classification

As described in the results section, we performed binary classification using the outcomes from correlation tests with and without the ground-truth units. A set of quantile thresholds ranged from the 0^th to 100^th quantile (figure 4(b)) of the two distributions of GT-Rs and test-Rs. Note that in rare cases, the best-R of the correlation test with ground-truth unit was not from the ground-truth unit. For such an outcome, we classified it as FP if it passed the thresholds, or TN if it did not. Then the F-score was calculated as follows:

(a) Illustration of how ground-truth surrogates were used to calculate True Positives (TP), False Negatives (FN), True Negatives (TN) and False Positives (FP) to determine the F-score for performance evaluation. (b) Example GT-R and test-R distributions, with an indication of the range of values for the two quantiles Q1 and Q2. (c) Performance matrix for one example network. For each pair of Q1 and Q2, the algorithm, detailed in A, was executed to determine the corresponding F-score. The green arrow indicates the maximum of the F-score.

Here we used 0.5 for β, as precision weighs more than recall in the test. The optimal thresholds could be identified when the F-score reached its maximum.