Repository for the preprint "Deep Generative Model for the Dual-Objective Inverse Design of Metal Complexes."
tmcinvdes will provide code and data from our research around Junction Tree Variational Autoencoder (JT-VAE) models that can generate metal ligands for transition metal complexes (TMCs). TMCs are used in industrial catalytic processes, anticancer therapies, and energy transformations. By labeling ligands with DFT-calculated target properties, conditional generative models can be trained and harnessed, optimizing metal ligands directionally in the target property space to discover novel TMCs by inverse design.
The code used to train the JT-VAE models and generate ligands are found at: JT-VAE-tmcinvdes.
Follow the environment setup instructions to install all the dependencies of the code in this repository. The code relies on a local download of the tmQMg-L repository and must therefore be cloned into a local folder.
Contains the code used to create the JT-VAE training sets.
Contains the code to assemble TMCs from ligands.
Contains the ORCA input file and parser scripts to label the generated ligands with the DFT-calculated properties of their homoleptic TMCs.
Contains the SA score code and the code for excluding outliers.
For a description of the workflow structure see the detailed workflow.
If you find our work useful, please cite our article:
@article{Strandgaard:2022:ChemRxiv,
author = {Magnus Strandgaard, Trond Linjordet, Hannes Kneiding, Arron Burnage, Ainara Nova, Jan Halborg Jensen, David Balcells}
title = {Deep Generative Model for the Dual-Objective Inverse Design of Metal Complexes},
journal = {ChemRxiv},
volume = {preprint},
doi = {10.26434/chemrxiv-2024-mzs7b},
year = {2024},
}
This repo will be gradually updated with code and data as the preprint goes through review.