XSMILES - JupyterLab example notebooks

Examples of pipelines from models & explanations to visualizations.

Available notebooks in notebooks/:

• Visualizing Gasteiger Charges (Simple example): notebooks/atom_attributions_gasteiger_charges.ipynb

  • Comparing LogD and bioconcetration factor attributions (Loading attributions from JSON): TBD.
  • Comparing LogP attributions from different methods (from ML Models to Attributions and Visualization): notebooks/smiles_attributions_for_logp.ipynb

Please Cite

If you use XSMILES, the use cases, its code, or the generated explanations, please cite our article:

https://jcheminf.biomedcentral.com/articles/10.1186/s13321-022-00673-w

Heberle, H., Zhao, L., Schmidt, S. et al. XSMILES: interactive visualization for molecules, SMILES and XAI attribution scores. J Cheminform 15, 2 (2023). https://doi.org/10.1186/s13321-022-00673-w

BibTeX @article{Heberle2023XSMILES, author={Heberle, Henry and Zhao, Linlin and Schmidt, Sebastian and Wolf, Thomas and Heinrich, Julian}, title={XSMILES: interactive visualization for molecules, SMILES and XAI attribution scores}, journal={Journal of Cheminformatics}, year={2023}, month={Jan}, day={06}, volume={15}, number={1}, pages={2}, abstract={Explainable artificial intelligence (XAI) methods have shown increasing applicability in chemistry. In this context, visualization techniques can highlight regions of a molecule to reveal their influence over a predicted property. For this purpose, some XAI techniques calculate attribution scores associated with tokens of SMILES strings or with atoms of a molecule. While an association of a score with an atom can be directly visually represented on a molecule diagram, scores computed for SMILES non-atom tokens cannot. For instance, a substring [N+] contains 3 non-atom tokens, i.e., [, {\$}{\$}+{\$}{\$}, and ], and their attributions, depending on the model, are not necessarily revealing an influence of the nitrogen atom over the predicted property; for that reason, it is not possible to represent the scores on a molecule diagram. Moreover, SMILES's notation is complex, foregrounding the need for techniques to facilitate the analysis of explanations associated with their tokens.}, issn={1758-2946}, doi={10.1186/s13321-022-00673-w}, url={https://doi.org/10.1186/s13321-022-00673-w} }

XSMILES for Javascript, KNIME, and How to use it

• XSMILES main project

• XSMILES for JupyterLab

How to run the notebook

Step 1 - Install general dependencies and XSMILES

Create a new virtual environment and install the dependencies defined in requirements.txt:

```bash

the code has been tested with Python 3.7, it's a dependency from CDDD

python3.7 -m venv .venvxsmilesusecases source ./.venvxsmilesusecases/bin/activate # path to the created environment pip3 install -r requirements.txt ```

Step 2 - Install CDDD

An unofficial package for CDDD is available in this repository: cddd-1.2.2-py3.none.any.whl. We packed CDDD scripts and the CDDD default_model into a single package to use in the notebook more easily, as well as to use with our Substitution method (attributor.py). Please check the smiles_attributions notebook to see how to we use the package and import the CDDD default model. We created this package because in certain environments, Google Drive may be blocked by firewalls.

bash pip install cddd-1.2.2-py3.none.any.whl

Make sure tensorboard==1.13.1 and tensorflow==1.13.2 were installed correctly through requirements.txt, CDDD depends on them, as well as on python <= 3.7.

You can use XSMILES for JupyterLab with newer versions of python. This dependency on Python 3.7 is here only for the CDDD model to work.

Step 3 - Run JupyterLab

Run JupyterLab and choose a notebook to explore:

bash jupyter lab notebooks

Notes

XSMILES from .whl file

If you don't want to install XSMILES from pipy (requirements.txt), you can install the .whl file available here

bash pip install xsmiles-0.2.1.dev0-py2.py3-none-any.whl

Internet connection is a requirement

The plugin will download RDkit MinimalLib when the JupyterLab notebook is loaded.