Machine-Learning Prediction of Metal-Organic Framework Guest Accessibility from Linker and Metal Chemistry

Area of Science:

• Materials Science
• Chemistry
• Crystallography

Background:

• Metal-organic frameworks (MOFs) are synthesized using metal ions and organic linkers, which dictate their structure and porosity.
• Selecting optimal metal-linker combinations for MOF synthesis is challenging due to the vast number of possibilities.
• Guest accessibility is a critical property of MOFs for applications in separation and catalysis.

Purpose of the Study:

• To develop a predictive model for MOF guest accessibility based on constituent metal and linker identities.
• To simplify the component selection process for experimental MOF synthesis.
• To provide guidance for designing MOFs with specific porosity characteristics.

Main Methods:

• Decomposition of experimental 3D MOF structures from the Cambridge Structural Database into metal and linker components.
• Machine learning models were trained to correlate chemical properties of metals and linkers with MOF porosity.
• Pore dimensions were classified into four ranges, and three sequential models were developed for prediction.

Main Results:

• A predictive model achieving 80.5% certainty in forecasting MOF guest accessibility solely from metal and linker identity was established.
• The relationship between component chemistry and MOF porosity was successfully learned.
• The developed dataset and predictive models are publicly available for use.

Conclusions:

• The study demonstrates that MOF guest accessibility can be reliably predicted from its fundamental building blocks.
• The predictive models offer a valuable tool for prioritizing component selection in exploratory MOF synthesis.
• This work facilitates the rational design of MOFs for targeted separation and catalysis applications.