Fig. 1.

Comparison between classical and machine-learnt (LE) encoding schemes for LSTM-based models of peptide-HLA-II interaction. The y-axis shows the area under the receiver operating characteristic curve, AUC, for the model predictions on the validation dataset (Val AUC). The x-axis shows the number of training cycles or epochs. a shows the performances of models trained on HLA-DRB1*15:01 data and b shows the model performances for HLA-DRB1*13:01 data