Integrating peptides' sequence and energy of contact residues information improves prediction of peptide and HLA-I binding with unknown alleles

Table 1 Prediction Results on Benchmark Dataset

Allele	ANNBM	ARB	SMM	NetMHC	Other methods	Peptides
A*0101	0.977	0.964	0.98	0.982	0.955	1157
A*0201	0.951	0.934	0.952	0.957	0.922	3089
A*0202	0.891	0.875	0.899	0.9	0.793	1447
A*0203	0.911	0.884	0.916	0.921	0.788	1443
A*0206	0.906	0.872	0.914	0.927	0.735	1437
A*0301	0.932	0.908	0.94	0.937	0.851	2094
A*1101	0.945	0.918	0.948	0.951	0.869	1985
A*2402	0.826	0.718	0.78	0.825	0.77	197
A*2601	0.950	0.907	0.931	0.956	0.736	672
A*2902	0.907	0.755	0.911	0.935	0.597	160
A*3101	0.923	0.909	0.93	0.928	0.829	1869
A*3301	0.915	0.892	0.925	0.915	0.807	1140
A*6801	0.88	0.84	0.885	0.883	0.772	1141
A*6802	0.883	0.865	0.898	0.899	0.643	1434
B*0702	0.966	0.952	0.964	0.965	0.942	1262
B*0801	0.968	0.936	0.943	0.955	0.766	708
B*1501	0.939	0.9	0.952	0.941	0.816	978
B*1801	0.848	0.573	0.853	0.838	0.779	118
B*2705	0.957	0.915	0.94	0.938	0.926	969
B*3501	0.873	0.851	0.889	0.875	0.792	736
B*4002	0.858	0.541	0.842	0.754	0.775	118
B*4402	0.824	0.533	0.74	0.778	0.783	119
B*4403	0.791	0.461	0.77	0.763	0.698	119
B*5101	0.894	0.822	0.868	0.886	0.82	244
B*5301	0.886	0.871	0.882	0.899	0.861	254
B*5401	0.911	0.847	0.921	0.903	0.799	255
B*5701	0.96	0.428	0.871	0.826	0.767	59
B*5801	0.972	0.889	0.964	0.961	0.899	988
AVG	0.909	0.791	0.874	0.901	0.796

Table 1 summarizes the comparative results between our ANNBM and the methods in the Bjoern Peters work on the benchmark. We use the AUC (Area Under roc Curve) of 5-folds cross validation as the prediction evaluation criterion. In the table, the first column is the allele name, including 14 HLA-A class molecules and 14 HLA-B class molecules. The columns from 2 to 5 are the AUC value of 5-folds cross validation from the ANNBM、 ARB、 SMM and NetMHC respectively. In addition, ANNBM is also compared to other 16 online prediction methods including various outstanding classifiers like SVM (arbmatrix, bimas, hlaligand, hla_a2_smm, libscore, mappp, mhcpathway, mhcpred, multipred, netmhc, pepdist, predbalbc, predep, rankpep, svmhc and syfpeithi) and the best prediction value among them is listed in the column 6. The last column is the number of peptides binding to the corresponding the HLA molecule.

ISSN: 1471-2105