High Structural Error Rates in "Computation-Ready" MOF Databases Discovered by Checking Metal Oxidation States

Area of Science:

• Materials Science
• Computational Chemistry
• Crystallography

Background:

• Computation-ready databases are crucial for high-throughput screening (HTS) and machine learning in materials discovery.
• The structural integrity of metal-organic framework (MOF) databases is largely unquantified, potentially impacting HTS and ML model reliability.
• Existing MOF databases may contain chemically invalid structures, leading to inaccurate computational screening results.

Purpose of the Study:

• To introduce MOSAEC, a novel algorithm for detecting chemically invalid MOF structures based on metal oxidation states.
• To quantify the prevalence of structural errors in leading MOF databases and HTS studies.
• To improve the reliability of computational materials discovery using MOFs.

Main Methods:

• Development of the MOSAEC algorithm to identify chemically invalid MOF structures by analyzing metal oxidation states.
• Manual validation of MOSAEC against 14,796 MOF structures from the CoRE database.
• Examination of over 1.9 million structures across 14 leading MOF databases and analysis of structures from 8 HTS studies.

Main Results:

• MOSAEC achieved 96% accuracy in detecting chemically invalid MOF structures during manual validation.
• Structural error rates exceeding 40% were found in most of the 14 leading MOF databases analyzed.
• Analysis revealed that 52% of top-performing MOF candidates from recent HTS studies were chemically invalid.

Conclusions:

• The MOSAEC algorithm effectively identifies chemically invalid MOF structures, enhancing data quality for computational screening.
• A significant proportion of MOF structures in leading databases and HTS studies contain errors, necessitating data validation.
• Improving the structural fidelity of MOF databases is critical for reliable materials discovery via HTS and machine learning.