Deep Learning Reveals Key Predictor of Thermal Conductivity in Covalent Organic Frameworks

This repository contains all the code necessary to reproduce the results presented in:

Thakolkaran et al., Deep learning reveals key predictors of thermal conductivity in covalent organic frameworks, Digital Discovery (2025). Paper link

---

Acknowledgements

All the code shown here is adapted from the MOFTransformer/PMTransformer repository:

• Park, H., Kang, Y., & Kim, J. (2023). Enhancing Structure–Property Relationships in Porous Materials through Transfer Learning and Cross-Material Few-Shot Learning. ACS Applied Materials & Interfaces, 15(48), 56375–56385. https://doi.org/10.1021/acsami.3c10323
• Repository: MOFTransformer.

We use publicly available hypothetical COF datasets from:

• Mercado, R., Fu, R., Yakutovich, A. V., Talirz, L., Haranczyk, M., & Smit, B. (2018). In Silico design of 2D and 3D covalent organic frameworks for methane storage applications. Chemistry of Materials, 30(15), 5069–5086. https://doi.org/10.1021/acs.chemmater.8b01425
  • Dataset link
  • We provide the subset of 2,471 2D COF cif files (1x1x1 unit cells) used from this dataset in this repository.
• Lan, Y., Han, X., Tong, M., Huang, H., Yang, Q., Liu, D., Zhao, X., & Zhong, C. (2018). Materials genomics methods for high-throughput construction of COFs and targeted synthesis. Nature Communications, 9(1). https://doi.org/10.1038/s41467-018-07720-x
  • Dataset link

---

Usage

The following file tree lists all files included in this repository.

main-folder/
│
├── data/
│   ├── charge/              # Dataset of TC of COFs with electrostatic interactions
│   ├── mercado-UC-cif/      # CIF files for the Mercado dataset
│   ├── multilayer-TC/       # Dataset of TC of selected multilayer COFs
│   ├── pSED/                # pSED data of selected COFs
│   ├── size-convergence/    # Dataset of TC of selected COFs used for size convergence study
│   ├── vacf/                # Dataset of VACF of selected COFs
│   ├── vdos/                # Dataset of VDOS of selected COFs
│   └── TC-data.csv          # COF dataset with thermal conductivity and dangling mass ratio values
│
├── drivers/
│   ├── preprocess.py        # Prepares the the files for PMTransformer 
│   ├── finetune.py          # Fine-tunes the pre-trained PMTransformer to predict the thermal conductivity
│   ├── predict.py           # Predict thermal conductivity using the fine-tuned PMTransformer
│   ├── visualize.py         # Visualize attention maps for specific COF files
│   ├── config.py            # Contains all configurations and hyperparameters.
│   └── environment.yml      # Conda environment file
│
├── visualize_results.ipynb  # Jupyter notebook to visualize data

To use this repository, follow these steps:

1. Install the dependencies using the provided environment file:

   conda env create -f environment.yml
   conda activate your_environment_name

2. Download the pre-trained PMTransformer model:

   moftransformer download pretrain_model

3. Preprocess the COF cif files and corresponding property labels:

   python preprocess.py

4. Fine-tune the pre-trained model using thermal conductivity (TC) data:

   python finetune.py

5. Predict and save the TC values for train/test/validation sets:

   python predict.py

6. Visualize attention maps for individual COFs:

   python visualize.py