RESUMEN
Background: Glioblastoma (GBM), one of the most prevalent brain tumor types, is correlated with an extremely poor prognosis. The extracellular matrix (ECM) genes could activate many crucial pathways that facilitate tumor development. This study aims to provide online models to predict GBM survival by ECM genes. Methods: The associations of ECM genes with the prognosis of GBM were analyzed, and the significant prognosis-related genes were used to develop the ECM index in the CGGA dataset. Furthermore, the ECM index was then validated on three datasets, namely, GSE16011, TCGA-GBM, and GSE83300. The prognosis difference, differentially expressed genes, and potential drugs were obtained. Multiple machine learning methods were selected to construct the model to predict the survival status of GBM patients at 6, 12, 18, 24, 30, and 36 months after diagnosis. Results: Five ECM gene signatures (AEBP1, F3, FLNC, IGFBP2, and LDHA) were recognized to be associated with the prognosis. GBM patients were divided into high- and low-ECM index groups with significantly different overall survival rates in four datasets. High-ECM index patients exhibited a worse prognosis than low-ECM index patients. Four small molecules (podophyllotoxin, lasalocid, MG-262, and nystatin) that might reduce GBM development were predicted by the Cmap dataset. In the independent dataset (GSE83300), the maximum values of prediction accuracy at 6, 12, 18, 24, 30, and 36 months were 0.878, 0.769, 0.748, 0.720, 0.705, and 0.868, respectively. These machine learning models were provided on a publicly accessible, open-source website (https://ospg.shinyapps.io/OSPG/). Conclusion: In summary, our findings indicated that ECM genes were prognostic indicators for patient survival. This study provided an online server for the prediction of survival curves of GBM patients.