Gradient boosted trees based predictor for MHC Class I epitope binding prediction. MHCBoost is also part of BERT, a very powerful deimmunization workflow.
How to install
Python 3.7 is a requirement! Support with 3.6 or lower is experimental! If you do get any errors about utf-8 you're likely not running MHCBoost with python 3.7.
Please also make sure that you're also installing MHCBoost using python 3.7 in a new virtual environment! Anaconda may help you with this. If you do get any errors about missing modules you're likely not installing this tool correctly. We tested the release thoroughly - it works perfectly fine!
If you're still having trouble with the setup, please write an e-mail to firstname.lastname@example.org . We're happy to help!
git clone https://github.com/igemsoftware2018/Team_Tuebingen_MHCBoost
python setup.py install
The CLI - Command Line Interface
> mhc-1 -p, --dataset_to_predict_path <arg> file to perform prediction on OR peptide sequence -o, --predicted_dataset_path <arg> filepath to save the predicted output file to -a, --allele <arg> allele to perform prediction on -l, --peptide_length <arg> epitope peptide length - usually 9 optional -t, --training_dataset_path <arg> file for classifier training optional -s, --silent suppresses learning output
Simply provide the answers to the questions asked by our tool.
Alternatively, provide input parameters when starting the tool.
> mhc-1 -p examples/example_input.txt -o /home/mypc/Desktop/output.txt -a A*02:01 -l 9
MHCBoost supports 65 alleles.
MHCBoost has an 5-fold crossvalidated average AUC of 0.899 on the IEDB dataset. The performance on each allele was compared to the state of the art NetMHCPan. Please refer to Results
Team iGEM 2018 Tübingen