Predicting MHC class I epitopes in large datasets

BMC Bioinformatics

Table 3 Prediction accuracies for the dataset containing only strong binders and clear non-binders

	Allele Name	Binders	Non-Binders	Total	DynaPredPOS	NetMHC	SVMHC	YKW
		*(n)*	*(n)*	*(n)*	*AUC*	*AUC*	*AUC*	*AUC*
1	A*0101	34	284	318	0.96	1.00	0.96	0.97
2	A*0201	549	503	1052	0.97	0.99	0.97	0.95
3	A*0202	290	267	557	0.98	1.00	0.97	0.97
4	A*0203	273	255	528	0.97	0.99	0.96	0.96
5	A*0206	216	371	587	0.97	0.99	0.96	0.95
6	A*0301	123	332	455	0.93	1.00	0.94	0.95
7	A*1101	228	269	497	0.95	1.00	0.94	0.97
8	A*2402	69	272	341	0.88	0.84	0.84	0.85
9	A*2601	15	256	271	0.97	1.00	0.94	0.97
10	A*3101	114	349	463	0.96	1.00	0.96	0.97
11	A*3301	36	620	656	0.95	1.00	0.94	0.93
12	A*6801	155	235	390	0.95	1.00	0.94	0.94
13	A*6802	95	440	535	0.96	1.00	0.97	0.94
14	B*0702	45	161	206	0.95	1.00	0.96	0.93
	Average AUC				0.95	0.99	0.95	0.95

Results from Dataset S, in which only strong binders (less than 10 nM binding affinity) and very clear non-binders (greater than 10,000 nM binding affinity) were included. Alleles included in Dataset F, which had fewer than 200 binders and non-binders in total in Dataset S, were no longer included in the analysis. The average AUC for each method is included at the bottom of each column.

ISSN: 1471-2105