Unlocking the potential of spatial transcriptomics

Paola Malsot’s academic path reflects her curiosity-driven approach to science. After completing a Bachelor’s degree in life sciences at EPFL, she pursued a Master’s in theoretical physics from EPFL, completing her thesis at ENS Paris. “I was drawn to life sciences for its impact on the medical sector, but I also loved the theoretical aspects of physics,” she explains. A pivotal experience working as a data scientist in the lab of virology and genetics at EPFL gave her the opportunity to combine both and led her to the ETH AI Center in late 2023.

Her research focuses on spatial transcriptomics, a novel molecular biology technique that maps gene expression patterns within tissue samples while preserving spatial context. This allows researchers to study how tumours interact with surrounding immune cells – a crucial step toward designing more effective cancer treatments. However, the technology presents significant challenges, “notably in terms of data processing, analysis and integration with other omics modalities,” Paola says. Her work involves developing AI-driven methods to identify and correct artefacts, helping to unlock the full potential of this powerful tool. While AI has already made significant strides in Natural Language Processing, applying it to biological data remains a challenge. “Large language models work so well on text because language is inherently structured and cleaned by humans,” she notes. “Biological data, on the other hand, is noisy and complex. We’re still in the early stages, but AI has the potential to transform our understanding of biology in the years to come.”

Paola is co-supervised by professors Gunnar Rätsch and Valentina Boeva, while also being part of the AI Center, which offers her a diverse and enriching research community. “The two labs have different research cultures, and it took time to adapt”, she says. “But now, I see it as an advantage – it gives me the best of both worlds.”

Truly creative AI for scientific discovery

Riccardo De Santi joined the ETH AI Center as a doctoral student in December 2023. His research spans three different labs, working under the supervision of professors Andreas Krause, Niao He and Kjell Jorner. Just like Paola, he found it challenging at first to align the different research cultures and “languages” spoken in these labs. “But I’m very happy now,” he emphasises.

Although he is still at an early stage of his research career, Riccardo’s long-term goal is clear: “I want to create a research identity that is uniquely mine,” he says. His work is at the interface between computer science and digital chemistry. In particular, he is developing AI algorithms that can assist exploration and optimisation over immense design spaces, as in drug discovery or material design. Riccardo envisions AI models that go beyond recombining existing knowledge – he is working on developing truly creative AI. Current generative models, like large language models, excel at generating compositions of known elements but struggle to produce genuinely novel insights. In one of his recent works, Riccardo shows how to guide generative models to maximise a mathematical notion of “surprise” to guide AI models into unexplored spaces. By doing so, these models are designed to actively seek out the unknown, making them powerful tools for accelerating scientific breakthroughs.

This methodology could significantly enhance the discovery of new catalysts, drugs and materials with specific properties. However, Riccardo sees applications beyond chemistry – for example in architecture or industrial design, where they could spur entirely new solutions for optimising material use or improving structural efficiency. “We’re currently building the algorithmic foundations for a new class of AI methods,” he explains. “Right now, the first applications are expected to be in catalyst synthesis, but ultimately, this new class of AI tools could tackle a wide range of scientific and design challenges.”

With AI and scientific discovery accelerating at an unprecedented pace, researchers like Paola Malsot and Riccardo De Santi are paving the way for breakthroughs that could redefine AI-driven research. The ETH AI Center fellowship, supported by the Dieter Schwarz Foundation, provides an ideal framework for this pioneering work – fostering interdisciplinary collaboration and pushing the boundaries of artificial intelligence in science.

“As part of our long-term partnership with ETH Zurich, we also support talented AI researchers – to ensure a responsible digital transformation.”

Reinhold Geilsdörfer
Managing Director Dieter Schwarz Stiftung