When the model allows changing species prevalence (i.e. GAM, GLM, BRT, CART), a prevalence of 50% was used in model weights. Performance is shown in terms of classification rate (AUC, sensitivity and
Percentiles and biases of cutpoints selected from cumulative ROC curves with several criteria and computed parametrically: 10,000 simulated datasets parameterized for proportional odds and AUC1 = AUC2
The results are the average across all the mouse pairs of Dataset A, computed employing all the metrics of Table 2 and 3. In all instances our approach outperforms [23]. The full results are available
