| Sep 19, 2024 |
|
(Nanowerk Information) For greater than 100 years, scientists have been utilizing X-ray crystallography to find out the construction of crystalline supplies corresponding to metals, rocks, and ceramics.
|
|
This method works greatest when the crystal is undamaged, however in lots of circumstances, scientists have solely a powdered model of the fabric, which comprises random fragments of the crystal. This makes it more difficult to piece collectively the general construction.
|
|
MIT chemists have now give you a brand new generative AI mannequin that may make it a lot simpler to find out the constructions of those powdered crystals. The prediction mannequin may assist researchers characterize supplies to be used in batteries, magnets, and plenty of different functions.
|
|
“Structure is the first thing that you need to know for any material. It’s important for superconductivity, it’s important for magnets, it’s important for knowing what photovoltaic you created. It’s important for any application that you can think of which is materials-centric,” says Danna Freedman, the Frederick George Keyes Professor of Chemistry at MIT.
|
|
Freedman and Jure Leskovec, a professor of laptop science at Stanford College, are the senior authors of the brand new research, which seems within the Journal of the American Chemical Society (“Crystal Structure Determination from Powder Diffraction Patterns with Generative Machine Learning”). MIT graduate pupil Eric Riesel and Yale College undergraduate Tsach Mackey are the lead authors of the paper.
|
|
| MIT researchers have created a computational mannequin that may use powder X-ray crystallography knowledge to foretell the construction of crystalline supplies. (Picture: Eric Alan Riesel)
|
Distinctive patterns
|
|
Crystalline supplies, which embody metals and most different inorganic stable supplies, are product of lattices that encompass many equivalent, repeating models. These models could be considered “boxes” with a particular form and dimension, with atoms organized exactly inside them.
|
|
When X-rays are beamed at these lattices, they diffract off atoms with totally different angles and intensities, revealing details about the positions of the atoms and the bonds between them. Because the early 1900s, this method has been used to investigate supplies, together with organic molecules which have a crystalline construction, corresponding to DNA and a few proteins.
|
|
For supplies that exist solely as a powdered crystal, fixing these constructions turns into way more tough as a result of the fragments don’t carry the complete 3D construction of the unique crystal.
|
|
“The precise lattice still exists, because what we call a powder is really a collection of microcrystals. So, you have the same lattice as a large crystal, but they’re in a fully randomized orientation,” Freedman says.
|
|
For hundreds of those supplies, X-ray diffraction patterns exist however stay unsolved. To attempt to crack the constructions of those supplies, Freedman and her colleagues skilled a machine-learning mannequin on knowledge from a database known as the Supplies Challenge, which comprises greater than 150,000 supplies. First, they fed tens of hundreds of those supplies into an present mannequin that may simulate what the X-ray diffraction patterns would seem like. Then, they used these patterns to coach their AI mannequin, which they name Crystalyze, to foretell constructions primarily based on the X-ray patterns.
|
|
The mannequin breaks the method of predicting constructions into a number of subtasks. First, it determines the scale and form of the lattice “box” and which atoms will go into it. Then, it predicts the association of atoms throughout the field. For every diffraction sample, the mannequin generates a number of doable constructions, which could be examined by feeding the constructions right into a mannequin that determines diffraction patterns for a given construction.
|
|
“Our model is generative AI, meaning that it generates something that it hasn’t seen before, and that allows us to generate several different guesses,” Riesel says. “We can make a hundred guesses, and then we can predict what the powder pattern should look like for our guesses. And then if the input looks exactly like the output, then we know we got it right.”
|
Fixing unknown constructions
|
|
The researchers examined the mannequin on a number of thousand simulated diffraction patterns from the Supplies Challenge. In addition they examined it on greater than 100 experimental diffraction patterns from the RRUFF database, which comprises powdered X-ray diffraction knowledge for almost 14,000 pure crystalline minerals, that that they had held out of the coaching knowledge. On these knowledge, the mannequin was correct about 67 % of the time. Then, they started testing the mannequin on diffraction patterns that hadn’t been solved earlier than. These knowledge got here from the Powder Diffraction File, which comprises diffraction knowledge for greater than 400,000 solved and unsolved supplies.
|
|
Utilizing their mannequin, the researchers got here up with constructions for greater than 100 of those beforehand unsolved patterns. In addition they used their mannequin to find constructions for 3 supplies that Freedman’s lab created by forcing parts that don’t react at atmospheric stress to kind compounds underneath excessive stress. This strategy can be utilized to generate new supplies which have radically totally different crystal constructions and bodily properties, though their chemical composition is similar.
|
|
Graphite and diamond — each product of pure carbon — are examples of such supplies. The supplies that Freedman has developed, which every comprise bismuth and one different component, may very well be helpful within the design of latest supplies for everlasting magnets.
|
|
“We found a lot of new materials from existing data, and most importantly, solved three unknown structures from our lab that comprise the first new binary phases of those combinations of elements,” Freedman says.
|
|
With the ability to decide the constructions of powdered crystalline supplies may assist researchers working in almost any materials-related area, based on the MIT group, which has posted an internet interface for the mannequin at crystalyze.org.
|
