TCRbridge

TCRbridge predicts TCR-pMHC reactivity from AlphaFold3 output.

Installation

Setting up conda from yml

1. Clone the repo:

   git clone https://github.com/schumacherlab/TCRbridge.git

2. Navigate to the project directory:

   cd TCRbridge

3. Create conda environment:

   conda env create -f tcrbridge_env.yml
   conda activate tcrbridge_env

Alphafold3

Option1: Alphafold3 structure predictions using alphafoldserver

For smaller sets of TCR-pMHC complexes alphafoldserver.com should suffice. The webserver allows for up to 20 protein complex predictions per day.

Option2: Alphafold3 structure predictions using a local installation

Follow instructions from https://github.com/google-deepmind/alphafold3 for local installation.

AlphaBridge

TCRbridge builds upon AlphaBridge (commit 3be1a44). To setup AlphaBridge:

1. Clone AlphaBridge

cd TCRbridge
git clone https://github.com/PDB-REDO/AlphaBridge.git
cd AlphaBridge
git checkout 3be1a44

2. Inside /TCRbridge/AlphaBridge/define_interfaces.py adapt define_interfaces() function to only calculate the metrics for contact_threshold_list = [0.4].

Performance

We ran all AlphaFold3 structure predictions on a sinlge A100 GPU. Calculating the TCRbridge score from AlphaFold3 output can be done on any machine in minutes.

Running

0. Generate the input files for Alphafold3

   Input data requirements:

   • CSV file with columns: tcr_id, full_seq_reconstruct_alpha_aa, full_seq_reconstruct_beta_aa, epitope_aa
   • The TCR alpha and beta sequences must be full-length sequences
   • The epitope sequence must match the peptide presented by HLA-A2 (otherwise you will need to adapt the .py script)

   python tcrbridge/make_af3_job_requests.py input_data.csv af3_job_requests/af3_job_request.json

1. Generate the AF3 structures

   Option 1: Using alphafoldserver.com (recommended for <20 structures)

   • Upload the generated json file to alphafoldserver.com
   • Download and extract the results to your output directory

   Option 2: Using local AlphaFold3 installation

   # Adapt paths in scripts/predict_structure_with_af3.sh to match your installation
   sbatch scripts/predict_structure_with_af3.sh af3_job_requests/af3_job_request.json /your/af3_structures/output/path/

2. Calculate interfaces and TCRbridge score for each TCR-pMHC complex

   # The script will:
   # 1. Calculate interfaces using AlphaBridge
   # 2. Generate TCRbridge scores for each structure
   # 3. Save predictions to .csv
   bash scripts/predict_tcrbridge.sh /your/af3_structures/output/path/ TCRbridge_predictions.csv

   Output: TCRbridge_predictions.csv containing TCR IDs and their TCRbridge scores

Citing This Work

BioRxiv link here

DOI: https://doi.org/10.1101/2025.04.28.651095

License

Distributed under the Apache 2.0 License.