This repo is intended as a documentation of the simulation experiments conducted in the scientific paper: Design of metal-organic frameworks using deep dreaming approaches.

The scripts are provided as-is, and are not guaranteed to work without the required dependencies or with a different structure from that used in our work. If you have any questions, please contact the corresponding author of the article.

An environment.yml file is available for setting up a new Python environment. Most of the required dependencies are handled in this file. However, the group selfies library must be installed separately --- v1.0.0 was used to develop the code in this repo. It is also recommended to install PORMAKE using pip or WSL (if using Windows) --- v.0.2.0 was used in the code development, and will be required to generate (decode) MOFs into their corresponding .cif files.

The code is primarily built upon the following libraries: pytorch, numpy, pandas, selfies, group selfies, rdkit, scikit-learn, ase, PORMAKE

1. To train a deep dreaming model, go to the directory /train_models/ and run the train_model.ipynb jupyter notebook. It is recommended to train these models on a GPU.
2. Pre-trained deep dreaming models are available in the /train_models/mof_saved_models/ directory.
3. To visualise the parity plots for these models, run the /train_models/plot_performance.ipynb jupyter notebook.

1. Go to the /deep_dreaming_experiments/ directory and explore the jupyter notebooks for the property case studies explored in the main article above (pore volume, $c_p$, and $Q_{CO_2}$ and $S_{CO_2/N_2}$). The studies make use of the pretrained models avaialable in /train_models/mof_saved_models/
2. Some of these studies rely on optimising seed distributions, which are contained in the /deep_dreaming_experiments/seeds/ directory.
3. Deep dreaming results are stored in the /dream_results/ subdirectory for each case study, and are visualised within their respective jupyter notebooks.

1. Go to the /construct_MOFs/ directory and run the construct_dreamed_mofs.ipynb jupyter notebook.
2. To construct MOFs, you need the linker SMILES string (or the file path to a pre-generated building block XYZ file), the PORMAKE toplogy string, and the PORMAKE node string.