Research at the Tumor Profiler Center

Jacob et al., Nature Communications, 2026

High-grade serous ovarian carcinoma, the most common form of ovarian cancer, is considered particularly difficult to treat. Chemotherapy remains the standard treatment, even though it has long been known how differently these tumors behave from patient to patient – and even within one and the same tumor in the same patient.

As part of the TPC, the research team investigated whether comprehensive, multilayered tumor analysis could improve treatment decisions within a clinically meaningful time frame. In 62 patients from Zurich, Basel, and Liestal, blood, tumor tissue, and ascites fluid were analyzed using up to eleven different technologies – ranging from genomic analysis and protein measurements to functional tests in which tumor cells were directly exposed to drugs. The complete analysis was available within four weeks in each case.

The result: an interdisciplinary tumor board changed the hypothetical treatment recommendation for a large proportion of patients when the additional analyses were taken into account. In one subgroup, maintenance therapy adjusted in this way after chemotherapy – for example, the choice of a specific PARP inhibitor or the addition of other agents – was associated with longer survival, although this observation must be interpreted cautiously given the lack of randomization. The study also showed that chemotherapy itself changes the composition of tumor cells: actively dividing cells decrease, while the remaining cells become more diverse. Tumors with a complete doubling of their genome stood out in particular – they tended to recur earlier and responded differently to drugs in laboratory tests.

The study shows that comprehensive tumor profiling can be carried out within clinically usable time frames and can provide valuable additional information for treatment. The findings are already feeding into a follow-up interventional study, the “OV Precision Trial.”

Link to publication: Nature Communications (2026)

Boos et al., npj Precision Oncology, 2026

In advanced, EGFR-positive non-small cell lung cancer (NSCLC), EGFR inhibitors are part of standard treatment. However, how long these drugs remain effective varies greatly from patient to patient, and the reasons for this are not always clear.

The research team examined two large real-world datasets: a cohort from the University Hospital Zurich comprising 43 patients, and an international database with almost 2,000 affected individuals. The central question was whether additional genetic alterations already present alongside the EGFR mutation at the time of diagnosis influence how long an EGFR-directed therapy remains effective.

The results showed that additional alterations in the ERBB2, PIK3CA, and TP53 genes were associated with a significantly shorter duration of treatment effect – a pattern that held true for both newer and older EGFR inhibitors and in both patient groups studied. TP53 alterations were the most common, and certain gene segments particularly important for the gene’s function were linked to a short duration of effect.

These findings support performing a more comprehensive genetic workup already at the time of diagnosis. This could help identify early on which patients may respond to standard therapy for a shorter time – and who might benefit from closer monitoring or an adjusted treatment strategy.

Link to publication: npj Precision Oncology (2026)

Wenckstern et al., Nature, 2026

So-called spatial proteomics methods show where in the tissue which cells are located and which proteins they carry – important information about, for example, how tumor and immune cells interact with one another. One problem: each study typically uses its own selection of markers and its own measurement technique, which has so far made it difficult to compare data across different studies.

The research team therefore developed “Virtual Tissues” (VirTues), an AI foundation model trained on tissue images from more than 3,000 patients and over one hundred different protein markers. Unlike previous approaches, VirTues can flexibly handle different marker combinations and measurement techniques. The model can identify and classify individual cells, detect functional neighborhoods within tissue, and even estimate values for markers that were not directly measured in a given study.

The benefit was especially striking in triple-negative breast cancer: patterns derived from VirTues predicted which patients would respond to a combination of chemotherapy and immunotherapy, and in an independent patient group they reliably distinguished between higher and lower risk of relapse – with markedly greater accuracy than previously established biomarkers and standard clinical classifications.

This work shows how artificial intelligence can help harness the growing complexity of modern tissue analyses. Such a tool could help bring together data from many different studies and technologies and identify new biomarkers for personalized cancer treatment. The long-term goal is also for model-based predictions to be tested in clinical trials in the future.

Link to publication: Nature (2026)