ml-XAI

KO | EN

gitlite — search

#javascript #python #hacktoberfest #react #ai #typescript #llm #go #golang #android #machine-learning #rust #deep-learning #linux

★ 9 Open GitHub ↗

No description available.

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# WISP - (Workflow for Interpretability Scoring using matched molecular Pairs)

WISP enables the quantitative evaluation of explainability methods applied to machine learning models. Details of the workflow are described in the following papers:
* [When Machine Learning Models Learn Chemistry I: Quantifying Explainability with Matched Molecular Pairs](https://chemrxiv.org/engage/chemrxiv/article-details/68c94d769008f1a467e0f1a5)
* [When Machine Learning Models Learn Chemistry II: Applying WISP to Real-World Examples](https://chemrxiv.org/engage/chemrxiv/article-details/68bb381ea94eede154ed44f8)

**Try out WISP in your browser — no installation required:**

<a href="http://www.molecularxai-lab.com/" target="_blank">Launch the WISP Web App</a>

The web application provides an interactive interface to explore explainability methods and quantitatively evaluate their performance — no prior coding knowledge required.

# Requirements for the Python Version
* python=3.11.11
* scikit-learn==1.6.1
* chemprop==2.2.0
* torch==2.7.1
* mmpdb==2.1
* rdkit==2024.9.6
* shap==0.47.2

# Installation

WISP is directly installable from GitHub:

```
pip install git+https://github.com/kerjans/ml-XAI.git
```

Or, if you want to clone the repository for development:

```
git clone https://github.com/kerjans/ml-XAI.git
cd ml-XAI
pip install -e .
git checkout code-paper-version-1
```

# Usage

We have provided different [notebooks](https://github.com/kerjans/ml-XAI/tree/main/notebooks) to:
* [Reproduce the publications](https://github.com/kerjans/ml-XAI/blob/main/notebooks/Reproduce-Paper.ipynb)
* See how [WISP is being used](https://github.com/kerjans/ml-XAI/blob/main/notebooks/examples_wisp.ipynb) and
* Explore how the [atom attributor is used](https://github.com/kerjans/ml-XAI/blob/main/notebooks/examples_atom_attributor.ipynb) as a standalone tool.

In general, you can execute WISP by simply plugging in your dataset of interest:
```python
wisp = WISP('working-dir-path',
     'dataset-path',
     'ID-column-name',
     'smiles-column-name', 
     'property-of-interest-column-name',
     model_available=None)
```

# Citations
Please acknowledge this work by citing the following if you use the package:

[When Machine Learning Models Learn Chemistry I: Quantifying Explainability with Matched Molecular Pairs](https://chemrxiv.org/engage/chemrxiv/article-details/68c94d769008f1a467e0f1a5):
```
@misc{jans2025wisp1,
      title={When Machine Learning Models Learn Chemistry I: Quantifying Explainability with Matched Molecular Pairs}, 
      author={Kerrin Janssen, Jan M. Wollschläger, Jonny Proppe and Andreas H. Göller},
      year={2025},
      archivePrefix={ChemRxiv},
      url={https://doi.org/10.26434/chemrxiv-2025-d57kp}, 
}
```
[When Machine Learning Models Learn Chemistry II: Applying WISP to Real-World Examples](https://chemrxiv.org/engage/chemrxiv/article-details/68bb381ea94eede154ed44f8):
```
@misc{jans2025wisp2,
      title={When Machine Learning Models Learn Chemistry II: Applying WISP to Real-World Examples}, 
      author={Kerrin Janssen, Jan M. Wollschläger, Jonny Proppe and Andreas H. Göller},
      year={2025},
      archivePrefix={ChemRxiv},
      url={https://doi.org/10.26434/chemrxiv-2025-zdqgr}, 
}
```

ml-XAI

Explore Similar Repositories

Was this content helpful?

Recent Feedback

Download README