Problems in training binding predicition model.

[KiAkize]

• BERTMHC version: 0.1.0
• Python version: 3.7.11
• Operating System: linux
• Non-commercial research [Y/N]: Y

Q1: In the cli.py file, aff_criterion = nn.BCEWithLogitsLoss(). However, the paper said We use the standard mean squared error loss function for the binding model and the weighted cross-entropy loss function (weight 10 for the positive class) for the presentation model when training with SA data. The code does not match the description of the paper, how did you train the model to get the results on the BA data?

Q2: In five-fold cross-validation and LOMO experiments on BA data, the same data as NetMHCIIpan3.2, what data should be set as --eval, respectively? What metic should be set as --metric, respectively?