Machine learning based early prediction of Glioblastoma using gene expression

Abstract

Glioblastoma multiforme (GBM), is an invasive highly malignant fast-growing tumor in the central nervous system. GBM ranks among the highest mortality rate cancers globally, with a survival rate of 5-10% even with combination therapies. The introduction of molecular prognostic-diagnostic biomarkers for central nervous system tumors, including microRNAs in exosomes plays a significant role in the early detection of Glioblastoma. The diagnosis by experts using clinical biomarkers is time consuming and variable. A Prediction Diagnostic model was developed for GBM based on miRNA molecular biomarkers from exosomes using Machine Learning algorithms. It was hypothesized to accurately distinguish glioblastoma patients from healthy individuals using the gene expression, offering a promising diagnostic tool for early detection of the disease with the help of Machine Learning.

Data was collected from the NCBI ^[1] Gene Expression Omnibus site. Eight machine learning models were trained and compared using accuracy, precision, recall, confusion matrices, and AUC ROC curves.  Logistic Regression was found to be the best model based on the comparison and matching the expert diagnosis. The study matches the six overexpressed microRNAs in GBM (hsa-miR-4443, hsa-miR-422a, hsa-miR-494-3p, hsa-miR-502-5p, hsa-miR-520f-3p, and hsa-miR-549a). Primarily the following two expressed microRNAs hsa-miR-549a and hsa-miR-502-5p  plays a significant role in the prediction of prognosis in patients with tumors of glial origin. A group of genetically expressed miRNAs that may serve as reliable biomarkers for brain cancer were identified using machine learning, which represents a powerful tool in biomarker identification.