Comprehensive SI

Comprehensive SI#

Date: Feb 13, 2026

Useful links: Github Repository

mlproject is a package that hosts all the utility scripts to reproduce the results from our publication: A critical assessment of bonding descriptors for predicting materials properties.

To keep the publication concise, not all results are included. This repository provides a one-stop access point to all results, along with all code and data required to reproduce them. Results are organized as static HTML pages for easy navigation and are deployed on github pages.

Preprint available at https://doi.org/10.48550/arXiv.2602.12109.

The code has been written by A. Naik with contributions from Dr. Philipp Benner.

Installation#

Please follow the steps listed below to be able to successfully create a conda environment to use all the scripts associated with this project. Following the order is important to avoid running into dependencies conflicts.

  • Use the provided dependencies.yml to create a conda environment using following command conda env create -f dependencies.yml. This will create a environment with python v3.10 and required dependencies.

  • Activate the newly created conda environment conda activate ml_env. Then install SISSO++ with python bindings. Adjust the CXX and CC paths based on your systems conda env path in command below.

    git clone --recursive https://gitlab.com/sissopp_developers/sissopp.git ~/sissopp && \
cd ~/sissopp && \
git checkout v1.2.6 && \
mkdir build && \
cp cmake/toolchains/gnu_param_py.cmake build/ && \
echo "set(SISSO_BUILD_TESTS ON CACHE BOOL \"\")" >> build/gnu_param_py.cmake && \
./build_third_party.bash CXX=/opt/conda/envs/ml_env/bin/g++ CC=/opt/conda/envs/ml_env/bin/gcc -j4 && \
cd build && \
cmake -C gnu_param_py.cmake ../ && \
make -j4 && \
make install && \
make test

  • Open the .bashrc and set LD_LIBRARY_PATH environment variable. Do not forget to adjust the path of conda env as per your system.

    export LD_LIBRARY_PATH="/opt/conda/envs/ml_env/lib:$LD_LIBRARY_PATH"

  • Install the rest of python dependencies in the same conda environment.

    pip install arfs==3.0.0 mendeleev==1.1.0 feature-engine==1.9.3 && \
pip install --upgrade emmet-core --force-reinstall --no-deps && \
pip install modnet==0.4.5 && \
pip install lobsterpy==0.5.8 && \
pip install . && \
pip cache purge

Alternatively, one can simply use the docker image associated with this repository to have an working environment setup directly with all required dependencies. For this one can simply open the repository in github codespaces or with vscode. Note for using devcontainer with vscode, one needs docker installed on the system.

Results and API reference#

Refer the rendered website for accessing all the results of manuscript. It also consists of scripts to reproduce all the results of the publication and API reference for the codes used.

Notebooks#

Target data extraction#

The scripts in notebooks/targets are used to extract the target datasets. The provided notebooks includes comments where need and should self explanatory. Please refer the notebooks/targets/*/README.md therein if it exists first before executing the scripts.

ML scripts#

The scripts in notebooks/ml_scripts can be used to reproduce the results of the manuscript. Each subdirectory is named as per sections in the manuscript and includes comments to make it self explanotory and easy to follow.

Misc#

The script in notebooks/misc can be used to generate the correlation plots shown in the manuscript.

ML models#

To support reproducibility, the trained machine learning models and post-processed data used to generate the figures in the publication are made available via Zenodo at the following DOI:

DOI

Other important citations#

Code#

Data#

Contents