Fig. 1

Binding predictions for both peptide-HLA-I and HLA-II pairs from ViPR correlate with their reported assay results. a The log10(percent rank) of scored peptide-HLA-I allele pairs, versus their ViPR reported MHC binding assay result (either binary Negative/Positive or the scaled Negative/Positive-Low/Positive-Intermediate/Positive-High, based on the reported value). In total, there were 4445 peptide-HLA-I allele pairs in the ViPR dataset we obtained (see the “Methods” section). Black lines indicate median values. b A modified precision-recall analysis for the HLA-I binding prediction of ViPR data, in which we demonstrate the fraction of the true positives out of all called positives (precision, indicated by color) and the fraction of detected true positives out of all true positives (recall, indicated x-axis value) as a fraction of the log10(percent-rank) threshold (y-axis value). Red dashed line indicates an example of log10(percent-rank) threshold of − 1, corresponding to the line in a. Area under the precision-recall curve is indicated. c The log10(percent rank) of scored peptide-HLA-II allele pairs, versus their ViPR reported MHC binding assay result (Negative+Positive-Low/Positive-Intermediate/Positive-High, based on the reported value). In total, there were 259 peptide-HLA-II allele pairs. Black lines indicate median values. d A modified precision-recall analysis for the HLA-II binding prediction of ViPR data, in which we demonstrate the fraction of the true positives out of all called positives (precision, indicated by color) and the fraction of detected true positives out of all true positives (recall, indicated x-axis value) as a fraction of the log10(percent-rank) threshold (y-axis value). Red dashed line indicates an example of log10(percent-rank) threshold of − 1, corresponding to the line in c. Area under the precision-recall curve is indicated