Decoding breast cancer with AI and RNA: Interview with DDM Prof. Raphaëlle Luisier

1. Could you briefly introduce your lab and the main questions your research is trying to address?

I lead the AI and Data Science for RNA Biology Lab at the University of Bern and founded SATURNA, an international initiative uniting five laboratories across artificial intelligence, RNA biology and translational science. We develop AI methods that integrate sequencing, imaging and clinical data to understand how RNA dysfunction drives complex human disorders such as cancer and neurodegeneration. Our aim is to turn that knowledge into therapeutic targets and, ultimately, RNA-based treatment strategies.

2. RNA plays a central role in how our cells function, but what can it tell us about breast cancer?

RNA is far more than a messenger between DNA and proteins. It helps control when, where and how genes are used, including protein production, subcellular organisation and gene regulation. In cancer, disruptions to RNA processing, including splicing, can reveal malignant cell states, tumour heterogeneity and mechanisms linked to progression or treatment resistance. Because many RNA processes can potentially be modulated, they also offer promising routes towards new therapies.

3. Your research identified different RNA-splicing patterns in breast cancer. Why does RNA splicing matter, and what might these patterns tell us about a tumour?

Our study specialised a histopathology foundation model on invasive tumour tissue and combined routine H&E images with spatial transcriptomic data. The model detected recurrent tumour regions with distinct molecular identities in HER2-positive and triple-negative breast cancers, including an archetype enriched for alternative-splicing activity. A greater abundance of this splicing-associated archetype was linked to poorer survival, suggesting that RNA-processing states capture clinically relevant tumour heterogeneity. A preliminary ovarian-cancer analysis identified a related pattern, although larger independent cohorts are needed.

We want to learn which patients may benefit from existing treatments while investigating new RNA-targeted approaches. This close bench-to-model-to-clinic cycle is how AI can move beyond pattern recognition and help generate discoveries that matter for patients.

4. How can AI help connect what we see in tumour tissue with what is happening at the molecular level?

We train models on standard H&E pathology slides and align the visual patterns they learn with spatial molecular measurements. This allows AI to distinguish tumour regions that may look similar to the human eye but carry different biological programmes; in our study, these included RNA splicing, immune activity and TGF-beta signalling. The model is not a replacement for the pathologist. Its value is as a discovery and stratification tool that can direct attention to biologically important regions at scale.

5. How could this type of research eventually contribute to better prognosis or treatment strategies for patients?

Histopathology is already part of routine cancer care, so extracting additional molecular information from the same slides could eventually support patient stratification without requiring a complex new assay for every case. Mapping splicing-associated regions may also help identify patients or tumour compartments that could benefit from RNA-modulating approaches. This study remains at the discovery stage: it is not yet a clinical diagnostic or a validated treatment selector. The next priorities are independent validation and testing whether these archetypes predict treatment response.

6. What do you see as the most exciting next step at the intersection of AI, RNA biology and breast cancer research?

The next step is to connect discovery across scales, from RNA sequence and structure to cell states, tissue architecture and patient trajectories, and then test what matters with clinicians and experimental scientists. Through SATURNA, we bring together complementary expertise in AI, RNA biology and translational research to build models that explain context-specific RNA function, identify therapeutic targets and guide the design of RNA-based strategies. In cancer, we want to learn which patients may benefit from existing treatments while investigating new RNA-targeted approaches. This close bench-to-model-to-clinic cycle is how AI can move beyond pattern recognition and help generate discoveries that matter for patients.