Fig. 4

Validation of composite matrices. Top: Binding likelihood (Z-score) means and standard deviations (sd) of binders and non-binders calculated using the matrices built based on Ssc and Hc and different scoring methods for pocket selection are shown for SLA class I and II alleles. Z-score means and sd calculated using negative matrices, HLA matrices with the lowest overall pocket identity for each SLA allele, are also shown. Instances where the Z-scores of binders and non-binders were statistically different (p-value ≤ 0.05) using a Wilcoxon-Mann–Whitney test, are shown in gray. Bottom: Comparison of binding likelihood (expressed as Z-score) between matrices (PAM30-Ssc and Negative) shown as density estimates (smoothed histograms). Note that y-axes are differently scaled. Binders and non-binders were scored with PAM30-Ssc (for class I), PAM30-Hc (for class II) and Negative control matrices. 100,000 natural random 9-mers were scored with either PAM30-Ssc (class I) or PAM30-Hc (class II). The black line indicates the threshold at which a 9-mer is considered a potential binder (Z-score of 1.64). Ssc was not applied to SLA-DRB1*0201 because crystal structures are not available