Skip to main content

Table 3 Benchmarking MHC class I predictors on recently published binding affinity data (HPV16), see also Table 4 for allele-wise scores

From: USMPep: universal sequence models for major histocompatibility complex binding affinity prediction

allele

mean AUC ROC

overall AUC ROC

USMPep_FS_ens

0.824(3)

0.814(3)

USMPep_FS_sng

0.818(4)

0.808(4)

USMPep_LM_ens

0.831(4)

0.815(3)

USMPep_LM_sng

0.813(5)

0.802(4)

MHCFlurry

0.817(4)

0.809(4)

NetMHC 3.4

0.831(3)

0.794(3)

NetMHC 4.0

0.803(4)

0.780(3)

NetMHCpan 2.8

0.818(4)

0.792(3)

NetMHCpan 3.0

0.815(4)

0.787(3)

NetMHCpan 4.0

0.820(3)

0.792(4)

SMM

0.684(5)

0.695(4)

SMMPMBEC

0.722(6)

0.723(4)

Pickpocket 1.1

0.760(5)

0.708(4)

consensus

0.751(5)

0.766(4)

IEDB recommended

0.756(5)

0.772(4)

NetMHCcons 1.1

0.827(4)

0.799(3)

  1. Predictive performance is evaluated by AUC ROC (threshold for binders < 100 000nM) on single alleles and across all alleles (mean and overall). The scores for literature approaches were calculated based on peptide-wise predictions provided in [12]. Numbers in brackets in the table concisely denote the corresponding bootstrap confidence intervals. For instance, 0.824(3) stands for a mean AUC ROC of 0.824 ± 0.003