While we're mainly aware of artificial intelligence (AI) and algorithms for machine learning because they predict our buying behavior, put together travel routes for us or identify faces in our photos, the interdisciplinary team of scientists from the University of Warwick, the TU Berlin and the University of Luxembourg was looking for a new application for this key technology. As AI is establishing itself in research as a crucial tool for supporting scientific discoveries, the researchers set about developing an AI algorithm that, for example, gives scientists the option of using the desired chemical properties of a substance as a basis for determining the necessary structure.

This option could play a major role in developing new drugs and materials in particular. However, to achieve this, AI needs to be able to incorporate fundamental laws of physics and learn how to solve quantum mechanics equations such as the Schrödinger equation. Yet solving these and similar equations in the conventional way requires vast computing capacity and, what's more, months of computing time. "This is precisely where we typically find a bottleneck with the computational design of new, purpose-built molecules for medical and industrial applications," explains Professor Klaus-Robert Müller, Professor of Machine Learning at the TU Berlin. Müller points out that the newly developed algorithm does not need any more computing capacity than a laptop or cellphone to deliver accurate predictions within seconds. "This has been a joint three-year effort, which required computer science know-how to develop an artificial intelligence algorithm flexible enough to capture the shape and behavior of wave functions, but also chemistry and physics know-how to process and represent quantum chemical data in a form that is manageable for the algorithm," adds Dr. Reinhard Maurer from the Department of Chemistry at the University of Warwick. "This interdisciplinary work is important progress as it shows that AI methods can efficiently perform the most difficult aspects of quantum molecular simulations," Müller says. And Professor Alexandre Tkatchenko from the Department of Physics at the University of Luxembourg concludes: "This work enables a new level of compound design where both electronic and structural properties of a molecule can be tuned simultaneously to achieve desired application criteria."

Technische Universität Berlin - Machine Learning Group (10623 Berlin, Germany)
Website: https://www.ml.tu-berlin.de/menue/maschinelles_lernen/parameter/en/