AI helps better understand how genetic mutations affect our health

VUB research compares language technology with classical biomedical methods

How do small changes in our DNA influence how our bodies function? That is the central question in the PhD research of Konstantina Tzavella at the Vrije Universiteit Brussel (VUB). Within the interdisciplinary TumorScope project and the VUB/ULB (IB)² Interuniversity Institute of Brussels, she explored how artificial intelligence (AI) can help predict which genetic mutations contribute to diseases such as cancer.

Mutations (small alterations in our genetic material) are essential for evolution but can also cause serious illnesses. Yet, the impact of most mutations remains unknown. “Predicting their effect is one of the biggest challenges in genetics,” says Tzavella.

To unravel this mystery, she compared existing state-of-the-art methods with a new generation of AI models inspired by language technology, similar to how systems like ChatGPT learn to understand language. These so-called protein Language Models (pLMs) learn to infer the structure and function of proteins from the sequence of amino acids, their building blocks.

“Just as language models learn how words combine to create meaning, pLMs learn how amino acids interact within a protein,” Tzavella explains. “This makes it possible to predict how mutations alter a protein’s function — and whether they might cause disease.”

One particularly challenging aspect is epistasis, the complex interplay between multiple mutations that can lead to unexpected effects. Where traditional approaches often fall short, pLMs appear to capture these interactions more effectively. By combining such models with evolutionary information, Tzavella developed a new computational model that not only makes more accurate predictions but is also clinically applicable, for instance to identify mutations that drive cancer growth.

“Our results show that pLM-based methods are both powerful and flexible,” she says. “They can even generate new insights into genes we still know little about.”

Originally from Akrata, Greece, Tzavella combined her background in electrical engineering and computer science with biomedical expertise. Through her PhD, she contributes valuable knowledge to the future of personalized medicine, where AI and biology are becoming increasingly intertwined.

Read more