Machine learning and statistical analysis to identify determinants of survival in primary glioblastoma using genome wide expression studies Dissertation for the award of the degree "Doctor rerum naturalium" of the Georg-August-Universität Göttingen within the doctoral program Environmental Informatics (PEI) of the Georg-

Background Glioblastoma (GBM) is an aggressive brain cancer that typically results in death in the first 15 months after diagnosis. There have been limited advances in finding new treatments for GBM. In this study, we investigated molecular differences between patients with extremely short (≤ 9 months, Short term survivors, STS) and long survival (≥ 36 months, Long term survivors, LTS). Methods Patients were selected from an in-house cohort (GLIOTRAIN-cohort), using defined inclusion criteria (Karnofsky score > 70; age < 70 years old; Stupp protocol as first line treatment, IDH wild type), and a multi-omic analysis of LTS and STS GBM samples was performed. Results Transcriptomic analysis of tumour samples identified cilium gene signatures as enriched in LTS. Moreover, Immunohistochemical analysis confirmed the presence of cilia in the tumours of LTS. Notably, reverse phase protein array analysis (RPPA) demonstrated increased phosphorylated GAB1 (Y627), SRC (Y527), BCL2 (S70) and RAF (S338) protein expression in STS compared to LTS. Next, we identified 25 unique master regulators (MR) and 13 transcription factors (TFs) belonging to ontologies of integrin signalling and cell cycle to be upregulated in STS. Conclusion Overall, comparison of STS and LTS GBM patients, identifies novel biomarkers and potential actionable therapeutic targets for the management of GBM. Graphical abstract