A library for robust and data-efficient classification of RepSeq data

Install

• Requirements
  • Python >= 3.8

git clone https://github.com/hmirin/MotifBoost
cd MotifBoost
pip install .

• To use with python3.7, set LOKY_PICKLER=pickle5

• To run on M1 Mac (2022 Jun.)

  • Remove immuneML==2.1.1 and numba-scipy scipy==1.6.2 from setup.py
  • Install immuneML==2.1.1 from source
    • Remove h5py from setup.py
  • Install numba-scipy from git repository
    • Remove scipy==1.6.2 from setup.py in this repository
    • scipy==1.7.3 will be used

Usage

• This library provides an easy interface for repertoire classification problem:
  • Repertoire provides the repertoire dataset class
  • MotifBoostClassifier provies a scikit-learn like interface for our robust and data-efficient repertoire classification method.

Example x

from motifboost.repertoire import repertoire_dataset_loader
from motifboost.dataset.settings import emerson_classification_cohort_split as settings
from motifboost.methods.motif import MotifBoostClassifier

# Get array of Repertoire instances
repertoires = repertoire_dataset_loader(
            save_dir,
            experiment_id,
            filter_by_sample_id,
            filter_by_repertoire,
            multiprocess_mode=multiprocess_mode,
            save_memory=False
        )

# Create train / val datasets from Repertoire metadata 
# Get target variable from Repertoire metadata
# Helper functions are already defined for this dataset
train_repertoires = [r for r in repertoires if settings.get_split(r) == "train"]
val_repertoires = [r for r in repertoires if settings.get_split(r) == "test"]
data_labels = [r.info["person_id"] for r in val_repertoires]

# MotifBoostClassifier implements a sklearn like API.
clf = MotifBoostClassifier()
clf.fit(train_repertoires, [get_class(x) for x in train_repertoires])
clf.predict_proba(val_repertoires, [get_class(x) for x in val_repertoires])

Citation

See our paper

@article{10.3389/fimmu.2022.797640,
author={Katayama, Yotaro and Kobayashi, Tetsuya J.},   	 
title={Comparative Study of Repertoire Classification Methods Reveals Data Efficiency of k-mer Feature Extraction},      
journal={Frontiers in Immunology},      	
volume={13},          
year={2022}, 	  
doi={10.3389/fimmu.2022.797640},      
}

Examples / Reproducing the results of the paper

• See https://github.com/hmirin/motifboost-paper