Table 4. Performance of various potentials on selecting models generated in CASP5-8 experiments as collected by Rykunov and Fiser [37].

Scoring function	models only		native included
	Averagea	ranked 1b	Averagec	ranked 1d
QMEAN6	2.87	85	1.71	113
RWplus	2.97	57	1.78	106
RW	3.08	51	1.71	110
QMEANall_atom	3.59	74	1.71	119
QMEANSSE_agree	3.74	62	3.72	39
QMEANACC_agree	4.04	40	3.78	48
RF_CB_SRS_OD	4.16	61	2.08	110
RF_CB_OD	4.62	62	2	111
RF_HA_SRS	4.65	49	1.38	137
RF_CB_SRS	4.72	56	2.18	114
OPUS_CA	4.72	79	5.13	55
VSCOREcombined	4.79	53	2.2	117
QMEAN-pairwise	4.8	54	3.15	85
Rosetta	5.01	57	4.09	68
Dong-pair	5.01	58	6.32	4
RF_CB	5.06	52	2.46	106
VSCORE-pair	5.08	54	1.85	128
PROSAcombined	5.11	57	3.38	87
OPUS_PSP	5.39	54	2.99	118
RF_HA	5.44	62	2.78	112
DOPE	5.77	54	3.27	95
dFIRE	6.03	50	5.69	33
PROSA-pair	6.03	56	3.54	95
QMEAN-torsion	6.71	45	3.24	114
Shortle2006	6.85	35	1.79	129
Liang_geometric	6.88	44	2.48	114
QMEANsolvation	7.32	33	6.27	54
Shortle2005	7.73	42	3.39	109
Floudas-CM	7.75	38	7.05	42
Floudas-Ca	7.79	33	8.36	10
NAMD_1000	8.06	24	4.96	78
Melo-ANOLEA	9.62	19	5.19	86
PC2CA	9.75	19	5.06	85
Melo-NL	9.99	14	5.85	80
NAMD_1	11.91	5	10.98	24
Randome	9.72	13.9	10.1	8.3

^aThe average rank of lowest energy decoy according to GDT_TS score (over 143 decoy sets) in the absence of native structures.

^bThe number of sets when the best model was ranked as first, in the absence of native structures.

^cThe average rank of the lowest energy decoy in GDT_TS when native structures are present.

^dThe number of sets when the best model was ranked as first when native structures are present.

^eExpected random values were generated by picking a wining model fromthe decoy sets randomly. Average values over 1000 random trials are shown [37].