Machine Learning and Omics Data for Predicting Overall Survival in Glioblastoma Patients

Abstract

Glioblastoma represents one of the most aggressive and fatal brain tumors, with patients facing a median survival of just 12 to 18 months. To improve understanding and diagnosis, multiple types of omic data-such as genomic, epigenomic, tran-scriptomic, proteomic, and metabolomic data-can be integrated [1]. Furthermore, including non-omic data, such as demographic details, clinical histories, and brain imaging, is essential to fully comprehend the disease and optimize patient care. In this study, we introduce an innovative approach for survival prediction in glioblastoma patients using omic data. A dataset from cBioPortal (n=619 samples) with 12 unique features was analyzed, applying a range of machine learning models to assess predictive per-formance. Specifically, we tested seven algorithms for survival prediction: Random Survival Forest, Random Forest Regressor, Gradient Boosting Regressor, XGBRegressor, Survival Support Vector Machine (SVM), Ridge Regressor, and Lasso Regressor. The results yielded a c-index of 0.78 for both Random Survival Forest and Survival SVM, 0.79 for Random Forest Regressor, Ridge Regressor, and Lasso Regressor, 0.80 for XGBRegressor, and 0.81 for Gradient Boosting Regressor. This study highlights the effectiveness of integrating omic data with machine learning to predict the survival rate in glioblastoma patients.