SmartCIF: a context-aware multi-agent system for automated preprocessing and curation of MOF CIFs

Area of Science:

• Materials Science
• Computational Chemistry
• Crystallography

Background:

• Computational screening of metal-organic frameworks (MOFs) relies on crystallographic data.
• Current CIF preprocessing methods often use fixed parameters, ignoring crucial details from original reports.
• This can lead to inaccurate results in high-throughput screening.

Purpose of the Study:

• Introduce SmartCIF, a novel framework for context-aware CIF preprocessing.
• Integrate literature information and topology-based analysis for chemically informed decisions.
• Enhance the accuracy and reproducibility of MOF computational screening.

Main Methods:

• SmartCIF couples topology-based structural analysis with natural-language reasoning.
• It makes chemically informed decisions on retaining or removing CIF parts based on user objectives.
• The framework integrates information from original publications to guide preprocessing.

Main Results:

• Benchmarking against 65 MOFs for BET surface areas and 321 data points for CO2/N2 adsorption.
• SmartCIF reconciles geometric accessibility with original publications, avoiding pore-blocking and nonphysical results.
• Demonstrated improved accuracy and reliability in MOF property prediction.

Conclusions:

• CIF preprocessing is application-dependent and requires explicit, controllable assumptions.
• SmartCIF provides a flexible, assumption-aware paradigm for reproducible and interpretable high-throughput screening.
• This approach generalizes existing resources and supports large-scale simulations beyond adsorption.