Menu
See all NewsEngineering News
Research

Algorithm Could Allow Faster Adoption of Versatile Materials

A machine-learning algorithm instantly predicts electronic properties in metal-organic frameworks

The new database can act as a map to reveal which metal-organic frameworks have desired electronic properties.The new database can act as a map to reveal which metal-organic frameworks have desired electronic properties.

Metal-organic frameworks (MOFs) are a new family of materials that can be tuned for a wide range of applications in clean energy and sustainability. While it’s promising that tens of thousands of MOFs have been synthesized with millions more proposed, that soaring number makes it extremely difficult and time-consuming to figure out which MOF best fits which application.
 
New work from a team led by Northwestern Engineering researchers could help solve this issue and make it easier to discover MOFs that have desirable conductive properties. 

Justin Notestein“The algorithm we developed allowed us to search through the library of every experimentally synthesized MOF to date and discover several MOFs with rare semi-conducting properties that were previously overlooked in the published scientific literature,” said Andrew Rosen, a McCormick School of Engineering PhD candidate in chemical engineering who is the lead author of the corresponding paper.
 
After producing the first large-scale database of MOF electronic properties, the team used that catalogue to train a machine-learning algorithm to predict MOF electronic properties in a fraction of a second on a standard laptop computer from nothing more than a depiction of its chemical structure, instead of several hundreds of hours of computing time on a state-of-the-art supercomputer or weeks-to-months in the laboratory.
 
This means it’s now possible to quickly discover which MOF works for which application, instead of researchers trying to find a needle in a haystack.
 
“We anticipate that this will help scientists much more efficiently identify exciting MOFs with a user-defined set of electronic properties that is perfectly suited for their needs, accelerating the pace of materials discovery,” Rosen said. 

Randall SnurrThe paper “Machine Learning the Quantum-Chemical Properties of Metal–Organic Frameworks for Accelerated Materials Discovery” was published April 5 in the journal Matter. Randall Snurr, John G. Searle Professor and chair of the Department of Chemical and Biological Engineering, and Justin Notestein, professor of chemical and biological engineering and director of the Center for Catalysis and Surface Science, part of the Institute for Sustainability and Energy at Northwestern, are co-authors in collaboration with researchers at the University of Chicago, the University of Minnesota, and the University of Toronto.
 
Many of the most pressing challenges that society faces within the areas of clean energy and sustainability are a race against time. However, it can take well over a decade to discover promising new materials for applications in the energy sector, if a solution is even found at all.
 
This is partly due to how the conventional scientific discovery process is often driven by time-consuming, trial-and-error efforts that are inherently limited in scope. 

Andrew Rosen“MOFs are an exciting new class of materials to address these challenges, as they have modular structures that can be tailored on the atomic level to have the ideal set of chemical properties,” Rosen said. “When coupled with the ever-increasing power of machine learning, the new database of MOF properties presented in our work makes it possible to rapidly search through the enormous combinatorial space of MOF structures in a matter of seconds, bringing us much closer to discovering ‘holy grail’ energy-relevant materials long sought after by scientists.”
 
The researchers collected the chemical structures of roughly 100,000 experimentally synthesized MOF structures, from which they simulated the electronic properties of roughly 15,000 representative materials using theoretical models based on the fundamental principles of quantum mechanics. That data was coupled with recently developed machine learning algorithms that could look at a MOF structure and predict its electronic properties from nothing more than a diagram of its chemical structure.