Table 2.

Comparison of Partial Least Squares and Neural Net.

	PLS			Method 1			NetMHCII			NetMHCIIPan
	AROC		r2	AROC		r2	AROC		r2	AROC		r2
	SB	WB		SB	WB		SB	WB		SB	WB
DRB1*0101	0.713	0.579	0.541	0.838	0.645	0.796	0.848	0.691	0.811	0.835	0.647	0.753
DRB1*0301	0.675	0.610	0.476	0.987	0.954	0.996	0.958	0.882	0.966	0.841	0.602	0.736
DRB1*0401	0.690	0.537	0.491	0.986	0.956	0.995	0.951	0.845	0.945	0.778	0.631	0.636
DRB1*0404	0.695	0.559	0.595	0.986	0.961	0.995	0.940	0.845	0.954	0.854	0.630	0.769
DRB1*0405	0.702	0.577	0.527	0.985	0.966	0.996	0.927	0.846	0.947	0.809	0.588	0.682
DRB1*0701	0.729	0.612	0.559	0.987	0.958	0.997	0.965	0.893	0.963	0.879	0.716	0.801
DRB1*0802	0.776	0.602	0.587	0.990	0.980	0.997	0.979	0.880	0.973	0.841	0.550	0.770
DRB1*0901	0.659	0.532	0.403	0.988	0.961	0.997	0.969	0.899	0.956	0.813	0.576	0.673
DRB1*1101	0.681	0.565	0.550	0.981	0.957	0.996	0.968	0.893	0.969	0.855	0.594	0.787
DRB1*1302	0.600	0.521	0.441	0.978	0.830	0.997	0.981	0.837	0.965	0.806	0.579	0.759
DRB1*1501	0.656	0.552	0.494	0.987	0.960	0.995	0.940	0.795	0.945	0.768	0.544	0.667
DRB3*0101	0.595	0.510	0.451	0.983	0.932	0.996	0.956	0.872	0.935	0.879	0.613	0.737
DRB4*0101	0.724	0.667	0.604	0.987	0.966	0.997	0.686	0.942	0.976	0.892	0.621	0.795
DRB5*0101	0.727	0.607	0.553	0.985	0.958	0.997	0.960	0.884	0.965	0.872	0.649	0.789
Average	0.687	0.574	0.519	0.975	0.927	0.982	0.931	0.857	0.948	0.837	0.610	0.740

The performance of partial least squares (PLS) compared to the neural network regression base on amino acid principal components (NN PCAA) described with two neural network predictors based on substitution matrices. SB and WB columns are the area under the receiver operator curve (AROC) obtained by converting the continuous for the regression fit output to a categorical output SB = strong binder (< 50 nM) WB = weak binder (> 50 nM and <500 nM) and non-binder (> 500 nM). The r² is indicated is the metric for how well the particular predictor predicts the values in the training set.