New AI systems predict practical applications of newly synthesized materials

Thousands of new materials are created each year, but the application is not immediately clear, so many new materials do not reach their full potential. Researchers at the University of Toronto aim to tackle the use of artificial intelligence.

In research published in Natural Communicationa team led by Seyed Mohamad Moosavi, an applied science and engineering researcher, has introduced AI tools that can predict how well new materials will work in real-world scenarios.

The system focuses on a class of porous materials known as Metal Organic Frameworks (MOFs) with adjustable properties and a wide range of potential applications.

Moosavi points out that materials scientists have created more than 5,000 types of MOFs last year alone, highlighting the scale of the challenge.

“In material discovery, the typical question is, 'What is the best material for this application?'' says Moosavi, an assistant professor of chemical engineering and applied chemistry.

“We turned the question upside down and asked, 'What is the best application for this new material?' With so many materials being made every day, I want to shift my focus from “which material to make next.” “What kind of evaluation should I give next?”

MOF can be used to isolate Co, for example₂ It prevents carbon from reaching the atmosphere from other gases in the waste stream and contributes to climate change. It can also be used to supply drugs to specific areas of the body or to enhance the functionality of electronic devices.

It is often found that MOFs created for one purpose have ideal characteristics for completely different applications. Moosavi cites previous research just seven years after its creation, which originally synthesized for photocatalysts, found to be extremely effective in carbon capture.

The new AI-powered approach aims to reduce this time delay between discovery and deployment.

To achieve this, PhD student Sartaaj Khan developed a multimodal machine learning system trained with different kinds of data that are normally available immediately after synthesis. In particular, precursor chemicals used to create materials and their powder X-ray diffraction (PXRD) patterns.

“Multimodality is important,” says Khan. “Just as humans use different senses, models draw more complete pictures by combining different kinds of material data to understand the world, such as vision and language.”

AI systems can use multimodal pre-training strategies to obtain insight into material geometry and chemical environments, allowing accurate property predictions without the need for post-synthesis structural characterization. This accelerates the discovery process and helps researchers identify promising material before it is overlooked or shelved.

To test the model, the team conducted a “time travel” experiment. AI was trained on available material data prior to 2017 and subsequently sought an evaluation of the synthesized materials. This system has successfully achieved several materials (developed for other purposes at origin) as powerful candidates for carbon capture. Some of them are currently undergoing experimental verification in collaboration with the National Research Council of Canada.

Musabi plans to integrate AI into the U of T's Acceleration Consortium's Autonomous Driving Laboratory (SDL).

“SDLS automates the process of designing, synthesising and testing new materials,” he says.

“When one lab creates new material, our system can evaluate it and potentially rerout it to another lab equipped to fully assess its potential.