Flowchart of testing the PhAI neural network. Source: Science (2024). DOI: 10.1126/science.adn2777
As part of an effort to predict the structures of small molecules, a trio of chemists from the University of Copenhagen have developed an AI application that can be used to work out the phases of X-rays diffracted by crystals.
In a paper published in the journal Science, Anders Larsen, Toms Rekis and Anders Madsen explain how they developed their system and how well it worked during testing.
Over the past few years, chemists and computer scientists have been working together to create AI applications that can be used to assist chemists in a variety of applications, and because a lot of chemistry is done by trial and error, such applications are a natural fit.
As an example, AI applications have recently been developed to predict protein structures. In this new effort, researchers have used AI to create an application that can do the same for small molecules.
As the researchers point out, the current process for predicting the structure of specific small molecules involves bundling them together, converting them into solid crystals, and then shining a beam of X-rays at them. Electrons in the X-ray beam bounce around in specific patterns after hitting the crystal. By analyzing those patterns, chemists can work out the structure of the molecules that make up the crystal.
But there's a problem: While it's relatively easy to measure the intensity of the X-rays as they're being emitted, researchers can't measure their phase. So they have to guess, which often results in what they call a blurry diffraction pattern. In this new study, the trio of researchers have found a way to use AI to find the uniqueness of the pattern, even when it's blurry.
To create the AI app, which they called PhAI, the team used computer models to create millions of fake small-molecule structures and then calculated the fuzzy diffraction patterns produced by those poor crystal structures.
We then used that result to train an AI on the relationship between crystals and the blurred patterns it generated, which gave us an output of millions of possible molecules, along with the phase and intensity information we needed. We used that information to perform the final training.
Tests of the system showed that it could accurately predict the structures of 2,400 real small molecules whose structures were already known. The team plans to continue their work with the aim of extending PhAI's capabilities to molecules with 50 atoms and beyond.
Further information: Anders S. Larsen et al., “PhAI: A deep learning approach to solving the crystal phase problem,” Science (2024). DOI: 10.1126/science.adn2777
© 2024 Science X Network
Citation: PhAI—AI system figures out the phase of X-rays diffracted by crystals (August 9, 2024) Retrieved August 9, 2024 from https://phys.org/news/2024-08-phai-ai-figures-phase-rays.html
This document is subject to copyright. It may not be reproduced without written permission, except for fair dealing for the purposes of personal study or research. The content is provided for informational purposes only.
