ARTEMIS

Algorithm for Rare-variant Trait Evaluation using Model-based Inference and Selection

ARTEMIS is Python-based tool for performing Bayesian variable selection in a classification model for observations with binary features, applied primarily to rare-variant gene-level collapsing data from UK Biobank. Demonstrated on a synthetic binary-valued dataset.

Installation

To install this python package, first download (clone) the git repo. Creating a new conda environment is recommended, commands to do this are listed below.

conda create -n ARTEMIS python=3.10
conda activate ARTEMIS
pip install -e .

Note that one must be in the git root directory to install.

Examples

A typical workflow for performing variational inference is as follows:

from artemis import artemis

# Initialise the variational inference module
vb = artemis.VB(
    X, # binary genotype matrix
    gene_names, # name of genes corresponding to each column of the genotype matrix
    y # phenotype labels (should be a binary vector of length matching the number of rows of X)
)

# Run variational inference (n_rep=number of iterations)
vb.run(n_rep=3)

See the examples.ipynb notebook for progressively more sophisticated examples. Note that to use the pipeline (last example in the notebook), one must first generate synthetic data using generate_synthetic_pipeline_data.ipynb.

This is not required to get started.

Citation