Machine learning the quantum-chemical properties of metal–organic frameworks for accelerated materials discovery

TLDR

Metal–organic frameworks are modular and allow tunable properties, yet identifying optimal MOFs for specific applications remains challenging, and quantum‑mechanical property prediction via high‑throughput screening and machine learning has been underexplored. The study introduces the QMOF database of quantum‑chemical properties for over 14,000 MOFs and aims to use it to train machine‑learning models that rapidly identify MOFs with desired electronic structure characteristics. Machine‑learning models were trained on the QMOF database to predict theoretically computed band gaps, enabling fast screening of MOFs for low‑gap electronic properties. The approach identified several MOFs predicted to possess low band gaps, overcoming the typical insulating nature of most MOFs.

Abstract

The modular nature of metal–organic frameworks (MOFs) enables synthetic control over their physical and chemical properties, but it can be difficult to know which MOFs would be optimal for a given application. High-throughput computational screening and machine learning are promising routes to efficiently navigate the vast chemical space of MOFs but have rarely been used for the prediction of properties that need to be calculated by quantum mechanical methods. Here, we introduce the Quantum MOF (QMOF) database, a publicly available database of computed quantum-chemical properties for more than 14,000 experimentally synthesized MOFs. Throughout this study, we demonstrate how machine learning models trained on the QMOF database can be used to rapidly discover MOFs with targeted electronic structure properties, using the prediction of theoretically computed band gaps as a representative example. We conclude by highlighting several MOFs predicted to have low band gaps, a challenging task given the electronically insulating nature of most MOFs.