Strength in numbers: achieving greater accuracy in MHC-I binding prediction by combining the results from multiple prediction tools

Table 1 Performances of the individual prediction tools on the HLA-A*0201 peptides from the community binding resource.

Tool	A _ROC	0.99 Specificity			0.95 Specificity
		Rank¹	Sensitivity	Threshold²	Rank	Sensitivity	Threshold
ARB Matrix	0.935	4	0.188	2.190	2	0.601	42.950
NetMHC 2.0 ANN	0.932	1	0.286	153.000	1	0.611	920.000
SMM	0.922	2	0.201	38.092	4	0.543	454.865
Bimas	0.920	3	0.198	324.068	3	0.552	47.991
SYFPEITHI	0.885	8	0.170	27.000	5	0.421	24.000
Multipred ANN	0.884	9	0.140	5.820	7	0.373	5.560
NetMHC 2.0 Matrix	0.872	6	0.177	24.329	10	0.358	20.129
SVMHC MHCPEP	0.870	12	0.115	1.000	11	0.334	0.520
Logistic Regression	0.862	13	0.101	0.364	9	0.364	0.108
SVMHC SYFPEITHI	0.854	7	0.176	0.950	8	0.367	0.490
HLA Ligand	0.825	10	0.137	141.000	14	0.274	127.000
Rankpep	0.822	15	0.077	103.000	13	0.306	83.000
MHCPred (Interactions)	0.818	5	0.182	43.350	6	0.377	99.080
MHCPred (position only)	0.814	11	0.116	21.330	12	0.311	65.310
Multipred HMM	0.798	14	0.090	7.530	15	0.244	7.090
Predep	0.788	16	0.045	-6.000	16	0.217	-5.130

The predictive performance of each tool for the HLA-A*0201 community binding data is shown using two measures: A _ROC score, and sensitivity when specificity is 0.99 and 0.95. ¹Indicates how the sensitivity of each tool compares to that of the other tools at the indicated specificity; the tool with rank 1 has the highest sensitivity. ²The scoring threshold corresponding to the indicated specificity.