Using Machine Learning and Data Mining to Leverage Community Knowledge for the Engineering of Stable Metal-Organic...

Area of Science:

• Materials Science
• Computational Chemistry
• Chemical Engineering

Background:

• Metal-organic frameworks (MOFs) offer tunable active sites and porous structures beneficial for applications like gas separations and catalysis.
• Limited understanding of MOF stability hinders their practical implementation, particularly their resistance to thermal degradation and solvent removal.

Purpose of the Study:

• To overcome the limitations of MOF stability for practical applications.
• To develop predictive models for MOF stability based on their chemical and geometric structures.
• To identify strategies for enhancing the stability of MOFs, especially those containing 3d-transition metals.

Main Methods:

• Extracted stability data (solvent-removal and thermal degradation) from thousands of published manuscripts using natural language processing and image analysis.
• Analyzed structure-property relationships within a large dataset of MOFs.
• Trained machine learning models (Gaussian process, artificial neural network) using graph- and pore-structure-based representations to predict MOF stability.

Main Results:

• Obtained over 2000 solvent-removal stability measures and 3000 thermal degradation temperatures from nearly 4000 manuscripts.
• Identified limitations of existing heuristics for MOF stability prediction.
• Developed ML models capable of predicting MOF stability orders of magnitude faster than conventional methods.
• Gained insights into key features influencing MOF stability through ML model interpretation.

Conclusions:

• Machine learning models can accurately predict MOF stability, significantly accelerating materials discovery.
• The developed approach provides strategies for engineering increased stability into MOFs for catalytic applications.
• This work is expected to expedite the development of stable MOF materials for diverse practical applications.