Sources of gain: Decomposing performance in conditional average dose response estimation
_{C. Bockel-Rickermann, T. Vanderschueren, T. Verdonck, W. Verbeke}

This repository provides code for our manuscript "Sources of gain: Decomposing performance in conditional average dose response estimation".

In our manuscript, we evaluate the impacts of different data-generating processes on data-driven methodologies for conditional average dose response (CADR) estimation. We provide source code to reproduce our experiments, including data generators, performance evaluators, and learning methods.

Code author: C. Bockel-Rickermann (christopher.rickermann@kuleuven.be)

Repository structure

This repository is structured as follows:

src-of-gain/
    |- src/               # Core library
        |- data/            # Data generators
        |- methods/         # Treatment effect estimators
        |- utils/           # Performance evaluation and other utils
    |- scripts/           # Executables
        |- exp/             # Reproduce experiments
        |- figures/         # Reproduce figures
        |- tables/          # Reproduce tables
    |- data/              # Data files
    |- config/            # Paramters for data loading and hyperparameter tuning

For reproducing experiments on TCGA datasets, download the necessary covariate matrices from here and save the data/ folder to the repository.

Installation

All code provided was written for python 3.9.16. To execute the code, please install the necessary packages to a newly created virtual environment by running:

pip install -r requirements.txt
pip install .

Running experiments

All executables are in the scripts/ folder. To execute them, simply run:

python scripts/[folder]/[script]

All results (performance metrics and plots) are saved to dedicated folders in the repository during execution.

Citing

Please cite our paper and/or code as:

@Article{src_of_gain,
  author        = {Bockel-Rickermann, Christopher and Vanderschueren, Toon and Verdonck, Tim and Verbeke, Wouter},
  title         = {Sources of gain: {D}ecomposing challenges in conditional average dose response estimation},
  year          = {2024},
  month         = 04,
}