INTRODUCTION: Glioblastomas (GB) are the most malignant primary brain tumors. Median survival is estimated at 15 months, although there is great heterogeneity amongst patients. To classify these phenotypically diverse cases there has been extensive research into the molecular background of GB patients, including identification of frequent mutations. For example, over 50% of GB cases present alterations in expression of epidermal growth factor receptor (EGFR), with over-expression associated to more aggressive GB. Multiple mechanisms lead to EGFR pathway activation but existence of different variants, and incomplete understanding of affected signalling pathways, have limited the application of treatments targeting this protein.

AIM(S): The presented project aims to specifically cluster these EGFR-altered patients and further characterize the biological pathways that differentiate groups.

METHOD(S): The algorithm applies unsupervised machine learning methods to GB patients' clinical and transcriptomic data for patient clustering, while further clarification of the biological basis of differences is done using a Drosophila melanogaster GB model.

RESULTS: Our algorithm identifies subgroups with differential patterns in tumor-associated biological processes, including significant differences in stemness, cell-cell communication, inflammation, chromatin destabilization, and tumor expansion. Identified genes of interest are also associated to differences in survival in the Drosophila model.

CONCLUSIONS: Further characterization of these stratifying mechanisms can lead to better understanding of tumor progression and the differences in response to treatments, with a final goal of more personalized tumor management.

FINANCIAL SUPPORT: The project is funded by a FPI project from the Comunidad de Madrid - IND2023/BMD-28759