---

English | 简体中文

This repository provides an implementation of the inference pipeline of TcrDesign.

We also provide the weights and datasets used for TcrDesign. For details, please refer to Zenodo.

Main models

TcrDesign comprises two components: TcrDesign-B (the binding model) and TcrDesign-G (the generation model).

TcrDesign-B accurately predicts interactions between epitopes and TCRs, incorporating full information such as VJ gene usage and MHC, and achieves state-of-the-art performance.

TcrDesign-G effectively generates a large number of epitope-specific TCRs.

Installation

First, please download the repository git clone https://github.com/XSLiuLab/TcrDesign

conda create -n tcrdesign python=3.8.16 && conda activate tcrdesign
conda install numpy=1.23.5 pandas=1.5.3 scikit-learn=1.2.2 tqdm=4.65.0 editdistance Levenshtein
conda install pytorch==1.12.0 torchvision==0.13.0 torchaudio==0.12.0 cudatoolkit=11.3 -c pytorch
pip install matplotlib seaborn tensorboard transformers

Please download tcrdesign_weights.tar.gz​ from Zenodo, extract and place the contents in the 'weights' folder​.

Usage

TcrDesign-B

Example 1: binding prediction for single sample

For missing values, please use 'X'​ as a placeholder.

python TcrDesign/tcrdesign_B.py -mode single -alphav TRAV12-2 -alphaj X -alpha_cdr3 CAVRGTGRRALTF -betav TRBV6-6 -betaj X -beta_cdr3 CASSFATEAFF -epitope GLYDGMEHL -mhc HLA-A02:01 -cuda False

Example 2: binding prediction for batch samples

Prepare a tab-delimited file following the format of TcrDesign/example/Binding_batch_example.tsv​. Convert MHC data using TcrDesign/data/mhc_pseudo/mhc_all.dat​ as reference. Replace missing values with X​.

python TcrDesign/tcrdesign_B.py -mode batch -data_path TcrDesign/example/Binding_batch_example.tsv -cuda True

TcrDesign-G

Example 1: generation bCDR3s for a specific epitope

python TcrDesign/tcrdesign_G.py -mode beta -epitope GILGFVFTL -num 100 -maxLen 20 -cuda True

Example 2: generation aCDR3s for a specific epitope

python TcrDesign/tcrdesign_G.py -mode alpha -epitope GILGFVFTL -bcdr3 CASSIRSTYEQYF -num 100 -maxLen 20 -cuda True

Example 3: generation VJs for a specific CDR3

# aCDR3 for VJ
python TcrDesign/tcrdesign_G.py -mode vj -acdr3 CAVNQGAQKLVF -cuda True
# bCDR3 for VJ
python TcrDesign/tcrdesign_G.py -mode vj -bcdr3 CASSIRSTYEQYF -cuda True

TcrDesign

Example: generation full-length TCRs for a specific epitope

python TcrDesign/tcrdesign.py -epitope GILGFVFTL -mhc HLA-A02:01 -gen_beta_num 30 -gen_alpha_num 100 -cuda True

Citing TcrDesign

If you use TcrDesign in your research, please cite our paper.

---

Cancer Biology Group @ShanghaiTech

Research group led by Xue-Song Liu in ShanghaiTech University

‍