ERGO

ERGO is a deep learning based model for predicting TCR-peptide binding.

Check our web-tool at http://tcr.cs.biu.ac.il

Requirements

pytorch 1.4.0
numpy 1.18.1
scikit-learn 0.22.1

Model Training

The main module for training is ERGO.py. For training, run:

python ERGO.py train model_type database specific gpu --model_file=model.pt --train_data_file=train_data --test_data_file=test_data

where:

• model_type is the the type of TCR encoding, LSTM based with lstm or autoencoder based with ae
• database is the training database, McPAS-TCR with mcpas or VDJdb with vdjdb.
• gpu is cuda device to use (e.g. cuda:0), or cpu for CPU (but it might be way slower)
• --model_file is the file which the model is saved to after training.
• --train_data_file and --test_data_file are train and test data files, you can set them as auto for defaults.

Binding Prediction

If you are interested in prediction only and not interested in training ERGO models, It might be more convenient to use our web tool, available here. You can choose what model and training set to use, and get the binding score of given TCRs and peptides from a csv file.

Anyway you can also predict using the ERGO.py module. It is quite similar to training, run:

python ERGO.py predict model_type database specific gpu --model_file=model.pt --train_data_file=train_data --test_data_file=test_data

where:

• --model_file is the trained model file.
• --test_data_file is a csv file with TCR and peptide columns. See example file in the ERGO website.
• All other cmd parameters are similar to the training process.

Models

The trained models and some of the train/test datasets we used are stored in the models directory.

TCR Autoencoder

The autoencoder based model requires a pre-trained TCR-autoencoder. for training the TCR-autoencoder, go to the TCR-Autoencoder directory using

cd TCR_Autoencoder

and run:

python train_tcr_autoencoder.py BM_data_CDR3s device model_file.pt

when device is a CUDA GPU device (e.g. 'cuda:0') or 'cpu' for CPU device. The trained autoencoder will be saved in model_file as a pytorch model. You can use the already trained tcr_autoencoder.pt model instead.

References

1. Springer I, Besser H, Tickotsky-Moskovitz N, Dvorkin S and Louzoun Y (2020) Prediction of Specific TCR-Peptide Binding From Large Dictionaries of TCR-Peptide Pairs. Front. Immunol. 11:1803. doi: 10.3389/fimmu.2020.01803