Glioblastoma (GBM) is an aggressive tumor type known to recur after maximal safe surgical resection followed by concurrent radiation therapy (RT) and chemotherapy (temozolomide—TMZ), and adjuvant TMZ maintenance chemotherapy. It exhibits high intratumor heterogeneity within a single specimen, and thus clinical management remains a challenge due to its rapid progression and high recurrence rate. Machine learning algorithms are currently being implemented in biomarker discovery to develop accurate predictive models that can guide clinical decision making. Emerging evidence identifies metabolomics as a critical player in understanding tumor metabolism and progression. Machine learning computation models have been instrumental in GBM classification and biomarker discovery, as well as the evaluation of tumor staging. Metabolomic profiling of biogenic amines in the setting of surgery, chemoradiation, and understanding relapse also suggests a coordination between metabolic pathways and tumor stage. Many challenges in machine learning and metabolomics-based approaches for disease classification remain due to the dimensionality of datasets, as well as identifying more streamlined panels of metabolite biomarkers. The purpose of this review is to showcase the recent developments in the applications of machine learning in metabolomics as a promising approach to enhancing the biomarker discovery process for future classification and interpretation of patient response to therapies for GBM management in the clinical setting. It also presents the major challenges of implementing machine learning approaches in GBM management and its future directions.
Shih et al. (Sun,) studied this question.