Enabling efficient exploration of metal–organic frameworks in the Cambridge Structural Database

Abstract

A tutorial review for mining the ever growing number of metal–organic frameworks data in the Cambridge Structural Database, for MOF scientists of all backgrounds.

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

“…All starting structures were taken directly from the CSD by querying the corresponding CSD reference code (''refcode''), and free (i.e., unbound) solvents were automatically removed from the frameworks. We chose to take the initial structures directly from the CSD as a matter of consistency and so that we could make use of valuable CSD metadata 70 (e.g., unresolved atoms, charged structures) associated with each deposited crystal structure. From this set of experimental crystal structures, we constructed a smaller DFT-ready subset of 42,349 nondisordered MOF structures (''QMOF-42349'') after an extensive suite of automated fidelity checks, as summarized in Figure S1.…”

“…This process serves to filter out many problematic MOFs with omitted H atoms, fractional occupancies, deleted framework atoms, lone (i.e., unbonded) atoms, overlapping atoms, an improper number of charge-balancing ions, and other structural issues that have been discussed in several recent studies. [70][71][72][73][74][75][76] Of these 42,349 experimental crystal structures, a subset of materials with 300 atoms or fewer per primitive cell was considered such that high-throughput DFT calculations could be carried out in an efficient manner. Full structure relaxations (including cell volume and atomic positions) were carried out via a multistage workflow 61 (Table S2) at the PBE-D3(BJ) 30,77,78 level of theory with the Vienna ab initio Simulation Package.…”

Targeted classification of metal–organic frameworks in the Cambridge structural database (CSD)

“…Any additional hydrogen-related disorder was removed by using the ‘non-disordered’ filter in ConQuest, following the protocol described recently to differentiate between the ‘non-disordered’ filter and the non-disordered MOF subset. 78 A PLD of 2.8 Å, corresponding to the lowest σ of the hydrogen atom across different force fields, was used to eliminate structures with lower PLDs.…”

Material Informatics with PoreBlazer v4.0 and CSD MOF Database

“…57,000 structures. In the next step, we removed the non-bonded solvent molecules as well as those bonded to open metal sites using the CSD Python API, according to the procedure by Li et al 41 Calculations on these ca. 57,000 structures revealed that only ca.…”