MQAPRank: improved global protein model quality assessment by learning-to-rank

BMC Bioinformatics

Table 1 The performances of the MQAPRank and several leading methods on CASP12 dataset based on GDT_TS score

Method	Method Type	Best 150^a				Sel20^b
Method	Method Type	Diff^c↓	MCC^d↑	AUC^e↑	Loss^f↓	Diff↓	MCC↑	AUC↑	Loss↓
MQAPRank	quasi-clustering	5.17	0.87	0.98	6.91	5.76	0.41	0.93	7.18
MUfoldQA_C	clustering	5.51	0.84	0.98	7.46	3.82	0.15	0.96	0.82
Davis-consensus	clustering	6.78	0.83	0.98	7.68	5.61	0.00	0.78	15.56
ModFOLD6_cor	quasi-single	6.75	0.86	0.98	10.55	6.70	0.86	0.99	1.28
MUfoldQA_S	single	8.90	0.71	0.93	13.15	3.60	0.76	0.98	2.56

^aBest 150: the dataset comprised of the best 150 models submitted on a target according to the benchmark consensus method. ^bSelect 20: the dataset comprised of 20 models spanning the whole range of server model difficulty on each target. ^cDiff: The average difference between the predicted and GDT_TS scores. ^dMCC: Matthews correlation coefficient (the threshold is 50 GDT_TS). ^eAUC: The area under the ROC curve. ^fLoss: The loss in quality between the best available model and the predicted best model. Bold value indicates highest performance

ISSN: 1471-2105