Skip to main content

Table 5 Language model and MHC class I binding affinity prediction performance

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

Model

LM

Downstream (mean)

 

perpl.

acc.

AUC ROC

Spearman r

LM (protein)

39.3

0.083

0.90(2)

0.55(4)

LM (peptide)

13.4

0.206

0.89(2)

0.57(4)

From scratch

–

–

0.89(2)

0.55(3)

  1. Language model metrics perplexity (perpl.) and accuracy (acc.) were in all cases evaluated on peptide data. The downstream performance corresponds to an ensemble of 10 predictors trained on the MHCFlurry18 and evaluated on the IEDB16_I test set
Back to article page

BMC Bioinformatics

ISSN: 1471-2105

Contact us