Development of Conformation Independent Computational Models for the Early Recognition of Breast Cancer Resistance Protein Substrates

Table 3

Results of the calculation of the total and partial areas under ROC curve for the best individual model and the 2 best 2-model ensembles.

	Model 1	Ensemble 1	Ensemble 2
Training set
Total ROC curve AUC (95% CI)	0.7964 (0.7284–0.8643)	0.8503 (0.7917–0.9089)	0.9022* (0.8578–0.9466)
Partial ROC curve AUC (±SD)
From 1 to Sp = [1 to 0.70]	0.1612 (±0.0199)	0.1877 (±0.0199)	0.2218 (±0.0163)*
From 1 to Sp = [1 to 0.75]	0.1252 (±0.0171)	0.1459 (±0.0178)	0.1771 (±0.0145)*
From 1 to Sp = [1 to 0.80]	0.0917 (±0.0147)	0.1059 (±0.0149)	0.1337 (±0.0122)^†
Simulated 577-compound database
Total ROC curve AUC (95% CI)	0.7321 (0.6413–0.8229)	0.7357 (0.6418–0.8297)	0.7707 (0.6746–0.8669)
Partial ROC curve AUC (±SD)
From 1 to Sp = [1 to 0.70]	0.1035 (±0.0213)	0.1127 (±0.0223)	0.1421 (±0.0223)
From 1 to Sp = [1 to 0.75]	0.0708 (±0.0183)	0.0794 (±0.0189)	0.1075 (±0.0208)^†
From 1 to Sp = [1 to 0.80]	0.0458 (±0.0140)	0.0504 (±0.0148)	0.0765 (±0.0162)^†

∗The value is different from the best individual model (model 1) (𝑃<0.001).
^†The value is different from the best individual model (model 1) (𝑃<0.01).