MixTCRpred accurately predicts T-cell receptors (TCRs) recognizing several viral and cancer epitopes (peptides displayed on MHC molecules or pMHCs). Predictions are available for 146 pMHCs. Accurate predictions were achieved for 43 pMHCs that have more than 50 training TCRs. Here the paper describing MixTCRpred predictive performance and applications.
You can run MixTCRpred within your web browser via Google Colab by clicking on this link. This is a user-friendly and interactive way to analyze the specificity of your own TCR list. For more extensive analysis or if you prefer to use MixTCRpred offline, it is recommended to install it on your local machine.
The code was tested with Python 3.9, 3.10 and 3.11
Clone the GitHub repository and move to the MixTCRpred directory
git clone https://github.com/GfellerLab/MixTCRpred
cd MixTCRpred
(Recommended) Create a virtual environment
# For Unix/Mac OS users
python -m venv MixTCRpred_venv
source MixTCRpred_venv/bin/activate # to activate the virtual environment (MixTCRpred_venv)
# Windows users please refer to https://docs.python.org/3/library/venv.html to create and activate a virtual environment.
Install the required packages
pip install --upgrade pip
pip install -r requirements.txt
To test your installation, run the following command:
python MixTCRpred.py --help
or
python MixTCRpred.py --list_models
which will list all available MixTCRpred models.
python MixTCRpred.py --model A0201_GILGFVFTL --input ./test/test.csv --output ./test/output.csv
to predict which TCRs in the ./test/test.csv file are more likely to target the HLA-A*02:01,GILGFVFTL epitope.
Before running the predictor, MixTCRpred also performs a quality control of the input data, attempting to correct incorrect V,J genes and extrapolating the CDR1 and CDR2 sequences. You can find the fixes and other information in the [output_file]_logfile file.
#change
path_pretrained_models = './pretrained_models'
#to
path_pretrained_models = '/home/[...]/MixTCRpred/pretrained_models'
Next make an alias to the MixTCRpred.py file using the python version of the virtual enviroment:
# For Unix/Mac OS users
alias MixTCRpred='/home/[...]/MixTCRpred/MixTCRpred_venv/bin/python /home/[...]/MixTCRpred/MixTCRpred.py'
# you can make this alias permanent by adding it to your .bashrc file
python MixTCRpred.py --model [MixTCRpred_model_name] --input [input_TCR_file] --output [output_file]
Three arguments are required:
--model
or -m
[MixTCRpred_model_name].
The format is HLA_PeptideSequence (e.g. A0201_GILGFVFTL).
--input
or -i
[input_TCR_file].
csv file listing all the TCRs to test. See ./test/test.cvs for a reference input file. The columns order is not important.
CDR3 alpha and beta should not be longer than 20 amino acids.
Incomplete TCR entries are accepted, but the models will have lower predictive performance
--output
or -o
[output_file].
The name of the output file. It contains two extra columns than the input file: the MixTCRpred binding score and the %rank.
Additional and optional arguments are:
--list_models
. To list the 146 MixTCRpred models for which we can currently run predictions. Models with less than 50 training TCRs have low confidence
--batch_size
. The default batch size for testing is 1. If you have a large dataset of TCRs to test, increasing the batch_size can speed MixTCRpred up
--download model_name
. To download a specific pretrained MixTCRpred model
--download_all
. To download the 146 pretrained MixTCRpred models
--download_high
. To download the 43 high-confidence pretrained MixTCRpred models
In the GitHub repository we include only two MixTCRpred models (A0201_GILGFVFT and A0201_ELAGIGILTV). You can download the pretrained MixTCRpred models from our Zenodo dataset
To download a specific pretrained model (e.g. A0201_NLVPMVATV) run:
python MixTCRpred.py --download A0201_NLVPMVATV
To download all the 146 pretrained MixTCRpred models run:
python MixTCRpred.py --download_all
To download the high-confidence 43 models (more than 50 training TCRs) run:
python MixTCRpred.py --download_high
For scientific questions, please contact Giancarlo Croce or David Gfeller
For license-related questions, please contact Nadette Bulgin.
This project received funding from the European Union's Horizon 2020 research and innovation programme under the Marie Skłodowska-Curie grant agreement, No. 101027973, MT-PoINT project