GHP-MOFassemble

Official repository of the paper "GHP-MOFassemble: Diffusion modeling, high throughput screening, and molecular dynamics for rational discovery of novel metal-organic frameworks for carbon capture at scale".

Authors: Hyun Park, Xiaoli Yan, Ruijie Zhu & Eliu Huerta

This novel framework combines generative AI, graph modeling, and high performance computing to enable high-throughput generation of novel pcu MOF structures with DiffLinker-generated linkers and designated nodes. Here is our paper's link.

Prerequisite

A list of required python packages can be found in ghp-mof.ipynb.

Dataset

data/hMOF_CO2_info.csv contains MOF name, MOFid, MOFkey, and isotherm data of 137,652 hypothetical MOF (hMOF) structures.

Workflow

To see how this framework works, please execute cells in ghp-mof.ipynb, which involves the following steps:

1. High-performing MOF structures (with CO2 capacity larger than 2 mmol/g @ 0.1 bar) are selected from the hMOF database
2. The MOFids of these high-performing MOFs are parsed to yield the SMILES strings of MOF linkers
3. Matched Molecular Pair Algorithm (MMPA) implemented in RDKit is used to fragment the unique linkers into their corresponding molecular fragments
4. DiffLinker is then used to sample new MOF linkers with number of sampled atoms varying from 5 to 10
5. The generated linkers are assembled with one of three pre-selected nodes into MOFs in the pcu topology
6. The modified CGCNN model proposed in our previous work is used to infer the CO2 capacities of the AI-generated MOF structures

Example high-performing MOF structures

18 predicted high-performing MOF structures that passed the molecular dynamics simulation density change criteria (<1%) are included in the high_performing_MOF_cifs folder.

License

This computational framework is released under the CC BY 4.0 Licence.