Table 2. Independent benchmarking of global scoring with official CASP12 data.

			GDT_TS	LDDT	CAD(AA)	SG
Rank	Gr.Name	Gr.Model	AUC	AUC	AUC	AUC
1	ModFOLD6_rank	QA072_1	0.993	0.99	0.926	0.962
2	ModFOLD6_cor	QA360_1	0.995	0.988	0.885	0.949
3	ModFOLD6	QA201_1	0.994	0.988	0.878	0.944
4	qSVMQA	QA120_1	0.982	0.983	0.862	0.937
5	ProQ3	QA213_1	0.985	0.978	0.892	0.916
6	ProQ3_1_diso	QA095_1	0.982	0.978	0.891	0.922
7	ProQ3_1	QA302_1	0.981	0.977	0.889	0.917
8	ProQ2	QA203_1	0.944	0.971	0.921	0.932
9	MUfoldQA_S	QA334_1	0.977	0.968	0.898	0.913
10	MULTICOM-CLUSTER	QA287_1	0.956	0.968	0.893	0.921

The ability of methods to separate good models (accuracy score ≥ 50) from bad (<50) according to GDT_TS, LDDT, CAD and SG scores is evaluated using the Areas Under the Curve (AUC) (see http://predictioncenter.org/casp12/doc/presentations/CASP12_QA_AK.pdf). Only the top 10 methods are shown and the table is sorted using LDDT scores. The scores are calculated over all models for all targets (QA stage 1–select 20). The table is sorted by the LDDT AUC score. Data are from http://predictioncenter.org/casp12/qa_aucmcc.cgi. See also Supplementary Tables S5–10.