xML-workFlow:
An end-to-end eXplainable Machine Learning workFlow for rapid biomedical experimentation

 

xML-workFlow is a scikit-learn template created to provide an alternative to monolithic scientific codebase. We designed xML-workFlow with 3 principles in mind:

• Modularisation: Minimum monolithic and maximum modularised code
• Abstraction: Don't repeat yourself, i.e. components used repeatedly should be abstracted
• Tracability: Tracable training, prediction and explainability with stateful logging of execution code (@NOTE: not yet supported) and analyses' arterfacts

By embedding these 3 principles into xML-workFlow, we aim to enable biomedical researchers to:

• reduce technical overhead in initiating ML experiments
• rapidly iterate over ML models, data versions, feature engineering techniques, etc with minimal refactoring / restructuring
• maintain visibility and tracibility as ML experiments scale up

Schematics of xML-workFlow framework

Using xML-workFlow

xML-workFlow was created as a template repository, i.e. the fastest way to use xML-workFlow is via the Use this template option.

Citing xML-workFlow

If you use the xML-workFlow template for building machine learning pipelines in your research, please cite the latest version of xML-workFlow at:

@article{Tran2025,
  title = {xML-workFlow: an end-to-end explainable scikit-learn workflow for rapid biomedical experimentation},
  url = {https://arxiv.org/abs/2504.01356},
  DOI = {10.48550/arXiv.2504.01356},
  journal = {aRxiv},
  publisher = {aRxiv},
  author = {Tran, Khoa A. and Pearson, John V. and Waddell, Nicola},
  year = {2025},
  month = April,
  pages = {2504.01356}
}