Table 1.

Summary of the performance (in PCC and AUC) of different prediction methods on the IEDB dataset

	PCC				AUC
Length	8	9	10	11	8	9	10	11
Alleles *	38	118	63	37	38	118	63	37
allmer	0.717	0.717	0.744	0.706	0.895	0.884	0.882	0.888
nmer	0.524	0.702	0.672	0.488	0.775	0.875	0.845	0.775
Lmer	0.664	0.702	0.699	0.670	0.871	0.875	0.860	0.868

^*For each length, only alleles with >20 data points and >3 binders are considered for validation.

allmer is the method trained on peptides of all lengths; nmer refers to networks trained only on peptides of length n; L-mer refers to networks trained on 9-mers and applied to peptides of different length using the L-mer approximation. Note that the L-mer approximation for 9mer reduces to the nmer method.