Prediction of metal-organic frameworks with phase transition via machine learning

Project Structure Overview

Preprocessing Directory

The preprocessing section is dedicated to the initial data preparation stages for the Prediction of Phase Transitions in Metal-Organic Frameworks (PPTiM). It encompasses scripts crucial for downloading essential resources such as the QMOF Database, Zeo++, and MOFid, setting the foundation for further analysis and model training.

Models Directory

Within the models directory, you will find the core computational models developed for this project. This includes an autoencoder designed to reduce the dimensionality of feature data, alongside classifiers tasked with predicting phase transitions in MOFs, showcasing the project's approach to leveraging machine learning for material science.

Database Directory

The database directory houses the PPTiM Database, a meticulously curated collection of MOFs known to exhibit phase transitions. This database not only includes CIF files detailing the structural data of these MOFs but also features manually labeled data pertaining to their phase transitions, serving as a valuable resource for research in this domain.

Knn Prediction Directory

The knn_prediction directory contains the data of MOFs from the QMOF Database labaled by knn prediction, trained by PPTiM Database.

Order of execution

To navigate the project's computational pipeline effectively, follow the order of execution outlined below:

1. for_zeopp - Prepares data for Zeo++ computation.
2. cif2vec_qmof - Processes CIF files from the QMOF Database using the cif2vec approach.
3. cif2vec_main - Processes CIF files from the PPTiM Database using the cif2vec approach.
4. cif2vec_t_sol
5. pipeline - Orchestrates the overall data processing and model training pipeline.
6. Figures - Conducts figures for metrics.

Acknowledgments and Citations

For researchers utilizing the PPTiM Database or methodologies developed within this project, please reference the associated publications.

• G.V. Karsakov, V.P. Shirobokov, A. Kulakova, V.A. Milichko. "Prediction of Metal–Organic Frameworks with Phase Transition via Machine Learning", The Journal of Physical Chemistry Letters 2024 15 (11), 3089-3095 DOI: 10.1021/acs.jpclett.3c03639

• Under review: "Phase Change Metal-Organic Frameworks: Current state and Application"

Installation

To install the required dependencies for this project, run the following command:

conda env create -f environment.yml

Prerequisites:

• Python 3.10
• CUDA Toolkit 12.1 (for torch)

If you do not have the CUDA Toolkit installed, you can install the CPU version of PyTorch.

Download QMOF Database

You can download the QMOF Database from the following link: QMOF Database

After downloading the QMOF Database, you should extract the folders from archive and place 'relaxed_structures' in this project's root directory.

Contact

For further information, inquiries, or interest in collaboration you can reach the corresponding author's details are available Vladimir Shirobokov and Grigori Karsakov.

Licensing

This project and all associated code and data are distributed under the GPL-2.0 license