In an attempt to forecast the structure of tiny molecules, a team of chemists from the University of Copenhagen have created an artificial intelligence program that can be used to determine the phase of x-rays that crystals have diffracted.
Anders Larsen, Toms Rekis, and Anders Madsen detail how they designed this system and how well it performed in testing in their publication that was published in the journal Science.
Together, computer scientists and chemists have been working on AI applications for several years now that can help chemists with a variety of tasks. These kinds of applications make perfect sense because a lot of chemistry is done by trial and error.
An AI program for predicting protein shapes is one example that was just created. Using artificial intelligence, the researchers have developed a new program that can perform the same function for small molecules.
The present method of predicting the structure of a specific small molecule, as noted by the researchers, entails growing batches of them into solid crystals and subjecting them to x-ray beams. Following impact, the electrons in the x-ray beam bounce about in a specific manner that allows scientists to deduce the structure of the molecules that comprise the crystal.
The problem is that, although it is not too difficult to gauge the x-rays’ strength when they are fired, it is not possible to gauge their phase. As a result, they frequently have to make educated guesses, which frequently results in diffraction patterns that they refer to as fuzzy. The research team’s latest work demonstrates how to leverage artificial intelligence (AI) to identify patterns’ uniqueness even when they are hazy.
In order to develop their artificial intelligence program, PhAI, the researchers first generated millions of fictitious tiny molecule structures using computer models, and then computed the fuzzy diffraction patterns that resulted from their subpar crystal structures.
The relationship between the crystals and the generated fuzzy patterns was then taught to the AI using the results. They received results for millions of potential molecules as well as the phase and intensity data they need by doing this. For their last training, they employed that knowledge.
The system’s ability to predict the structure of 2,400 real small molecules whose structure was known was demonstrated during testing. In order to increase PhAI’s capabilities beyond 50 atom molecules, the research team intends to carry out more study.