How do you use artificial intelligence to simulate chemical processes? Philippe Schwaller has developed a program that has been named the best of its kind by an independent research group.
Enthusiastic, passionate, analytical – and yes, a bit nerdy. In terms of his appearance and gestures, the young man sitting in the Grosse Schanze park in Bern on this sunny afternoon and using his hands to try and explain his research comes across a bit like the character Q from the latest James Bond films. Someone who clearly enjoys his work and has been able to turn what he loves into a career. Or, to put it another way, to bring a sense of playfulness to the field of science.
Even as a secondary school pupil, the Freiburg native was interested in everything to do with engineering and natural sciences – no wonder, then, that he decided to study something that unites the two: materials sciences is essentially a combination of chemistry, physics and engineering sciences. During his year studying abroad in Manchester, the EPFL student used computer models to simulate materials, thus pursuing his interest in computer science, programming and machine learning. His skills in these fields are mainly self-taught "with online courses".
"As if atoms were letters, molecules were words and reactions were sentences"
As an intern at the IBM Research Centre, he dove deeper into the topic that would eventually win him the Prix Schläfli: linguistics models in chemistry. He further developed this approach while completing his PhD at the IBM Research Centre and the University of Bern. "The main theme of my doctoral thesis was how to take artificial intelligence models that were developed for human language and apply them to molecules and reactions," he explains. "It's as if atoms were my letters, molecules my words and reactions my sentences." In the same way that machine translation programs are fed with millions of sentences in order to translate text from one language to another, Philippe Schwaller fed the neural networks with chemical reactions. In future, this kind of approach could be used to automate chemical synthesis and significantly reduce the amount of time required for the discovery and production of new molecules.
Philippe Schwaller is not the first person to pursue this path. However, what makes his work so unique and has now earned him the Prix Schläfli is an aspect that, according to the experts, represents a breakthrough and has eclipsed the competition: Schwaller discovered that his models can identify the way in which atoms reconfigure themselves during reactions, something referred to as "atom mapping". Using these patterns, he has developed a quick, high-quality atom mapping tool – the RXNMapper. Many chemists have never heard of atom mapping, but it is essential for digital chemistry. Computerised atom mapping has been a subject of research since the 1970s because it is of fundamental importance for most computer-assisted tools for planning syntheses, can help make reaction simulations more efficient and makes reaction data more accessible and easier to interpret and search. Moreover, Schwaller has also managed to set a new standard: "An independent group of researchers has compared various approaches and selected my open-source RXNMapper as the best atom-mapping tool – even better than commercially available solutions," says the 31-year-old.
"I'm much better on the computer than in the lab"
At the moment, Schwaller is an Assistant Professor of Digital Chemistry at EPFL. "A dream," he says. "I can make my ideas a reality and let my creativity run wild." He wants to use his skills to do "something useful" and to help avoid potential failures in the lab. In fact, the lab is not really his preferred working environment. "I had several opportunities to conduct experiments and I witnessed for myself that things don't always go according to plan in the lab, and that many factors influence the results," he says. He then adds, laughing: "I'm much better on the computer than I am in the lab."
However, there is also a lot to his life away from the computer: the son of parents from both the French-speaking and German-speaking regions of Switzerland, he enjoys spending his (rare) free time hiking, camping, travelling and taking photos – or cooking with his partner. As to whether or not he experiments with molecular gastronomy? That is one question that will have to remain unanswered.
Using language models to facilitate chemical syntheses, improve the understanding of large earthquakes, decipher the fundamentals of cell biological processes, produce single photons for protected data transfers – the Swiss Academy of Sciences (SCNAT) is awarding the Prix Schläfli 2022 to the four most important insights of young researchers at Swiss universities. Luca Dal Zilio (Geosciences), Anna-Katharina Pfitzner (Biology), Philippe Schwaller (Chemistry) und Natasha Tomm (Physics) receive the prize for findings in their dissertations. The Prix Schläfli was first awarded as early as 1866.
Her work could help give data transfers more protection against being hacked: during her dissertation, Natasha Tomm (co-)developed a super-efficient source of individual photons.Image: Clemmens Spinnler
Small biological building blocks are her thing: biologist Anna-Katharina Pfitzner has researched a mechanism that is key to many processes in cell biology.Image: Anna-Katharina Pfitzner
Large earthquakes are once-in-a-century events with devastating consequences. Luca Dal Zilio has developed a model that describes the development of such events both temporally and geographically, and which could therefore become important for risk prevention.Image: Victoria Lasheras