RPEMHC

RPEMHC: improved prediction of MHC-peptide binding affinity by a deep learning approach based on residue-residue pair encoding

Requirements

• paddle == 2.3.2
• python == 3.7

RPE(residue-residue pair encoding)

python preprocess.py  ##before reproduction, several files need to be generated by preprocess.py  

Experiments

python train.py --mode cv   ##5-flod cross-validation on IEDB2016 
python train.py --mode lomo    ##Leave one molecule out on IEDB2016
python train.py --mode binary_test   ##independent test on Binary_test
python train.py --mode iC50_test   ##independent test on IC50_test
python train.py --mode epitope     ##independent test on T-cell epitope benchmark
python train.py --mode ablation   ##ablation experiment
python train.py --mode binding_core   ##binding core prediction
python train.py --mode seqlogo   ##generate data to draw seq2logo
python train.py --mode CD4   ##evaluation compared with NetMHCIIpan-4.0
python train.py --mode CD8   ##evaluation compared with NetMHCpan-4.1

trained model

Trained models are provided in https://pan.quark.cn/s/2985aa5a6138