[Preprint]. 2023 Jul 21:2023.07.05.547832. Originally published 2023 Jul 5. [Version 2] doi: 10.1101/2023.07.05.547832

Table 1.

Area under the ROC curve (AUC) value for protein model quality classes as a function of different scoring metrics.

Score^a	Binary classification^b		Multi-class classification^b
Score^a	Incorrect vs. High	Incorrect vs. Medium and High	Multi-class classification^b
interface pLDDT	1.00	0.93	0.88
model confidence	0.99	0.88	0.85
ipTM	0.99	0.87	0.85
pTM	0.99	0.89	0.85

Scoring methods. “interface pLDDT”: average pLDDT of antibody-antigen interface residues within 4 Å. Models without antibody-antigen interface contacts were assigned an interface pLDDT value of 30.

The ROC values of binary classification and multi-class classification were calculated using the R pROC⁵⁸ and multiROC⁵⁹ packages. Protein model quality class is assigned as per CAPRI criteria, into Incorrect (n=9,112), Acceptable (n=773), Medium (n=684) and High (n=156) classes.