For over 100 years, scientists have used X-ray crystallography to determine the structure of crystalline materials such as metals, rocks and ceramics.
The technique works best when the crystal is intact, but in many cases, scientists only have a powdered version of the material that contains random fragments of the crystal. That makes it harder to piece together the entire structure.
MIT chemists have developed a modern generative AI model that could make it much easier to determine the structures of these powdered crystals. The predictive model could facilitate researchers characterize materials for utilize in batteries, magnets, and many other applications.
“Structure is the first thing you need to know about any material. It’s important for superconductivity, it’s important for magnets, it’s important for knowing what photovoltaics you’ve created. It’s important for any application you can imagine that involves materials,” says Danna Freedman, the Frederick George Keyes Professor of Chemistry at MIT.
Freedman and Jure Leskovec, a professor of computer science at Stanford University, are lead authors of the modern study, which appears today in The paper’s lead authors are Eric Riesel, a senior at MIT, and Tsach Mackey, a first-year student at Yale University.
Unique designs
Crystalline materials, which include metals and most other inorganic solids, consist of lattices that are made up of many identical, repeating units. These units can be viewed as “boxes” of characteristic shape and size, with atoms arranged precisely within them.
When X-rays are shined onto these lattices, they bend the atoms at different angles and intensities, revealing information about the positions of the atoms and the bonds between them. Since the early 1900s, the technique has been used to analyze materials, including biological molecules with crystalline structures, such as DNA and some proteins.
For materials that exist solely in powdered crystal form, solving these structures becomes much more complex because the fragments do not contain the full three-dimensional structure of the original crystal.
“The exact lattice still exists because what we call the powder is actually a collection of microcrystals. So you have the same lattice as the large crystal, but they are in a completely random orientation,” Freedman says.
X-ray diffraction patterns exist for thousands of these materials, but they remain unresolved. To try to decipher the structures of these materials, Freedman and her colleagues trained a machine-learning model on data from a database called the Materials Project, which contains more than 150,000 materials. First, they fed tens of thousands of these materials into an existing model that can simulate what the X-ray diffraction patterns would look like. Then, they used those patterns to train their AI model, which they call Crystalyze, to predict structures from the X-ray patterns.
The model divides the structure prediction process into several subtasks. First, it determines the size and shape of the lattice “box” and which atoms will be in it. Then, it predicts the arrangement of the atoms in the box. For each diffraction pattern, the model generates several possible structures that can be tested by feeding the structures into the model, which determines the diffraction patterns for that structure.
“Our model is generative AI, meaning it generates something it hasn’t seen before, and that allows us to generate several different guesses,” Riesel says. “We can make a hundred guesses and then predict what the powder pattern for our guess should look like. And if the input looks exactly the same as the output, then we know we did it right.”
Solving unknown structures
The researchers tested the model on several thousand simulated diffraction patterns from the Materials Project. They also tested it on more than 100 experimental diffraction patterns from the RRUFF database, which contains powdered X-ray diffraction data for almost 14,000 natural crystalline minerals that they had separated from the training data. On that data, the model was true about 67 percent of the time. Next, they began testing the model on diffraction patterns that had not been previously resolved. Those data came from the Powder Diffraction File, which contains diffraction data for more than 400,000 resolved and unresolved materials.
Using their model, the researchers worked out structures for more than 100 of these previously unsolved patterns. They also used their model to discover structures for three materials that Freedman’s lab created by forcing elements that don’t react at atmospheric pressure to form compounds at high pressure. This approach can be used to generate modern materials that have radically different crystal structures and physical properties, even though their chemical compositions are the same.
Graphite and diamond—both made of pure carbon—are examples of such materials. The materials developed by Freedman, which contain bismuth and one other element, could be useful in designing modern materials for eternal magnets.
“We discovered many new materials from existing data and, most importantly, we solved three unknown structures from our lab that constitute the first new binary phases of these element combinations,” Freedman says.
According to the MIT team, which published a web interface for the model at: “The ability to determine the structures of powdered crystalline materials could help scientists working in almost any materials-related field.” crystallization.org.
The research was funded by the U.S. Department of Energy and the National Science Foundation.