GRATCR

GRATCR is a pre-trained deep learning model designed to generate T cell receptors (TCRs) for specified epitopes.

Requirements

GRATCR was trained on two GeForce RTX 3080 GPUs with 20GB memory using the PyTorch 1.10.1 framework and Python 3.9. To set up the environment and install the necessary dependencies, run the following commands:

conda create -n GRATCR
conda activate GRATCR
conda install -r requirements.txt

Model Pre-training

To pre-train the BERT, run:

accelerate launch BERT.py --data_path='./Data/merged/epitope.csv' --vocab_path='./Data/vocab/total-epitope.csv' --model_path="./model/bert_pretrain.pth"

To pre-train the GPT, run:

accelerate launch GPT.py --data_path='./Data/merged/beta.csv' --vocab_path='./Data/vocab/total-beta.csv' --model_path="./model/gpt_pretrain.pth"

Model Fine-tuning

To fine-tune the GRATCR, run:

accelerate launch GRA.py --data_path='./Data/MIRA/MIRA.csv' --tcr_vocab_path='./Data/vocab/total-beta.csv' --pep_vocab_path='./Data/vocab/total-epitope.csv' --model_path="./model/gra.pth" --bert_path="./model/bert_pretrain.pth" --gpt_path="./model/gpt_pretrain.pth" --mode='train'

TCR Generation

To generate TCRs for given epitopes, run:

python GRA.py --data_path='./Data/MIRA/MIRA.csv' --tcr_vocab_path='./Data/vocab/total-beta.csv' --pep_vocab_path='./Data/vocab/total-epitope.csv' --model_path="./model/gra.pth" --bert_path="./model/bert_pretrain.pth" --gpt_path="./model/gpt_pretrain.pth" --mode='generate' --result_path='./result.csv'

Data and Pre-trained Model Weights

Some of the data and pre-trained model weights are stored in a cloud drive for your convenience. You can download them and use them directly in your experiments.

Download the Data: Access the following link to download the necessary data files: Data Download Link
Download Pre-trained Model Weights: Access the following link to download the pre-trained model weights: Model Weights Download Link

After downloading, place the files in the appropriate directories as specified in the commands above.