Brain tumors aren't all the same — even within the same diagnosis, some grow faster than others. A new computational approach called Dynamic Quantum Clustering (DQC) analyzed gene activity patterns from 692 brain tumor samples and found that it could sort them into groups that matched standard clinical diagnoses 91% of the time.
The method, which works without needing doctors to label the data first, also identified a 554-gene subset that separated tumor types with 97% accuracy. One cluster was nearly all glioblastomas (the most aggressive type), with a 97% positive predictive value. Among low-grade gliomas, the analysis revealed three distinct subgroups with different survival outcomes, hinting at hidden biology that could matter for prognosis.
This is early work — the analysis used public RNA sequencing data, not new patient samples. The researchers caution that the method is still exploratory and not ready for clinical use. But it shows how machine learning that respects the natural shape of data might one day help doctors classify tumors more precisely and tailor treatment.