MatCreatioNN

Table of Contents

• Introduction
• Features
• Repository Structure
• Trained Models
• Datasets
• Usage
• Contributing
• License

Introduction

MatCreatioNN is an integrated computational framework for the design, evaluation, and cost estimation of Metal–Organic Frameworks (MOFs). The system incorporates:

• Reinforcement-learning MOF generation
• Crystal Graph Convolutional Neural Networks (CGCNNs)
• A multi-stage prediction funnel for stability, catalytic ability, cost, sustainability, and adsorption
• Automated XRD simulation, property prediction, and fitness scoring

This framework enables large-scale screening of both experimental MOFs and AI-generated MOFs for catalytic, adsorption, and economic applications.

Features

• Cost Estimation Dataset: A curated dataset assigning industrial cost values to each element in its commonly used form.
• Trained CGCNN Modules: Pre-trained neural network models for predicting MOF properties including stability, adsorption, selectivity, cost, and synthesis likelihood.
• Reinforcement-Learning MOF Generator: Constructs chemically valid MOFs using tokenized metal clusters and linker fragments.
• Multi-Stage Prediction Funnel: Efficient filtering using quantile-based thresholds, significantly reducing computational cost while preserving top-performing candidates.
• Full Reproducibility: All scripts, datasets, models, and preprocessing tools required to reproduce results in the associated publication.

Repository Structure

GitHub Repository: https://github.com/SatyaK-0/MatCreatioNN/tree/main

1. MachineLearningPredictionModels_PTHFiles/ – Contains the trained PyTorch .pth model files for all 13 CGCNN predictors used in the funnel/filtration system.
2. LogicProof/ – A script demonstrating that the ordering of funnel stages does not affect final outcomes when using quantile-based filtering.
3. SupportingPythonPrograms_Implementation_DataProcessing/ – Supporting scripts for:
   • Reinforcement-learning MOF generation
   • PyMatGen-based XRD simulation
   • Molecular cost calculator
   • Post-processing of funnel outputs
   • Normalization, visualization, and utility functions
4. cgcnn/ – Modified and bug-fixed CGCNN implementation updated for compatibility with modern PyTorch and Python versions.

Trained Models

This repository contains pre-trained neural-network models for all MOF-related predictions performed in the MatCreatioNN framework, including:

• Thermal and water stability predictors
• CO₂ adsorption and CO₂/H₂O selectivity models
• Available surface area predictor
• Synthesis likelihood classifier
• Elemental cost and sustainability regressors

Usage

To run the full pipeline:

1. Install Python ≥ 3.10 and PyTorch ≥ 2.0.
2. Clone the repository and install required dependencies.
3. Use the provided scripts in SupportingPythonPrograms_Implementation_DataProcessing/ to generate structures, compute descriptors, and evaluate funnel performance.

Each directory contains additional documentation or comments to guide usage.

Contributing

Contributions are welcome! To contribute:

1. Fork the repository.
2. Create a new branch for your feature or fix.
3. Commit your changes and push the branch.
4. Open a pull request with a clear explanation of your contribution.

License

This project is released under the MIT License unless otherwise specified.