Fig. 5
From: USMPep: universal sequence models for major histocompatibility complex binding affinity prediction
Performance of our MHC I prediction tools compared to MHCFlurry on single alleles. Spearman r was calculated for predictions on the Kim14Blind data for alleles with more than 25 quantitative measurements. The predictors were trained on Kim14BD2009. The alleles are ranked by the size of the corresponding training set (9528 peptides for rank 0 to 136 peptides with rank 52). No MHCFlurry predictors were provided for alleles HLA-B-2703, HLA-B-0803 and HLA-B-3801 with rank 45, 49 and 52