This episode of ByteSight features Dr. Irene Paassen, Postdoctoral Researcher in the Rubin Lab at the University of Bern. A molecular biologist by training, Irene did her doctoral research at the Princess Máxima Center for Pediatric Oncology in Utrecht, where she built patient-derived organoid models of malignant rhabdoid tumors, one of the rarest and most aggressive childhood cancers. Today she studies how cancer cell identity shifts over time and drives treatment resistance in advanced prostate cancer.

In this conversation, Manasi and Irene explore what it means to do research when only a few patients a year are diagnosed in an entire country. They discuss why pediatric diagnostics often go beyond genetics, why only around 8% of more than 50,000 genomically profiled patients in a Japanese cohort received a matched therapy, and why the field may need more than one reference genome to serve every patient.

About Dr. Irene Paassen

She is a molecular biologist specialized in cancer epigenetics. She has performed her doctoral research at the Prinses Maxima Centrum in Utrecht focusing on the molecular diversity of malignant rhabdoid tumors in children. This work deepend her curiosity to understand cancer cell trajectories and understand their adaptation to changing environment and treatment. Within her postdoctoral research at the University of Bern under supervision of Prof. Mark Rubin she is now studying the dynamics of lineage switching of aggressive prostate cancer.

Complexity in Simplicity

Irene’s thesis title describes the tumor she studied: a single gene lost, often before birth, yet several distinct subtypes growing out of it. She also sees it as a description of cancer research as a whole. The goal is simple, but the biology behind it is not, and the field is still missing pieces of the puzzle.

There’s a simple question we want to answer in cancer research: how can we kill cancer cells as quickly and efficiently as possible? But it’s such a complex picture for such a simple question.

She also explains why pediatric diagnostics work differently. In adult cancers, diagnosis rests largely on genetics and pathology. In childhood tumors, genetics may only tell you where to start. Subtyping a pediatric brain tumor, for example, often depends on DNA methylation profiling rather than mutations alone.

When the Whole Country Is Your Sample

Centralized care in the Netherlands gave Irene access to every pediatric case in the country. Even so, only around five patients a year were diagnosed with the tumor she studied. Her first biobank held 11 organoids across three subtypes, with one subtype never represented at all. The organoids supported drug screens that were valuable for the individual patients, but predicting treatment response for future patients was another matter.

Even in a centralized institution, there are just too few samples per year to build a robust model. You really need to work towards sharing data from different institutions to build such a robust model on rare cancers.

Asked what a truly global dataset would change, Irene points to epigenetics rather than genetics. Environmental factors and what children inherit from their parents shape the epigenome, and she expects those signals to differ fundamentally across populations.

What AI Can Realistically Do With Small Data

Irene sees two areas where AI already helps or soon will: faster, more reliable imaging and in silico drug discovery, validated afterwards in the lab. For rare cancers, the condition is appropriate training data. Because rare cancers will always produce few data points, she argues for transferring knowledge from cancer types where more data exists.

The sparsity in the data of a rare cancer will never change. So you either need the perfect scenario where everyone shares data, or you need to leverage knowledge from other cancer types.

The 8% Gap in Precision Medicine

With Mark Rubin, Irene wrote a News & Views piece in Nature Medicine on a Japanese study that followed more than 50,000 genomically profiled cancer patients. Only around 8% received a therapy guided by their genetic profile. She had expected closer to 20%.

Part of the gap is access: many mutations have no matching therapy yet. Part is the data itself. The study used a targeted panel of roughly 326 genes, and many newer therapies, from epigenetic drugs to immunotherapy, rest on more than genetics. Irene argues for looking at transcriptomics and other layers to understand what a mutation actually does.

The discussion then turns to the reference genome. Irene and Manasi agree that a reference built largely on one population will miss what is present in others.

Maybe we need to adjust our reference. Maybe we need to start having multiple reference genomes, based on who we have and from comparing data.

Final Takeaway

Rare cancers will never produce large datasets on their own. Closing the gap means sharing data across institutions and borders, looking beyond genetics, and building references that reflect the patients they are meant to serve. Irene remains optimistic: the field keeps uncovering new problems, but it also has the innovation to solve them.

I wouldn’t see it as, oh, there’s even more complexity coming. I think we also have enough innovation to tackle that complexity, to find more puzzle pieces for the bigger puzzle.