Machine learning-based identification of lower grade glioma stemness subtypes discriminates patient prognosis and drug response

Abstract

Glioma stem cells (GSCs) remodel their tumor microenvironment to sustain a supportive niche. Identification and stratification of stemness related characteristics in patients with glioma might aid in the diagnosis and treatment of the disease. In this study, we calculated the mRNA stemness index in bulk and single-cell RNA-sequencing datasets using machine learning methods and investigated the correlation between stemness and clinicopathological characteristics. A glioma stemness-associated score (GSScore) was constructed using multivariate Cox regression analysis. We also generated a GSC cell line derived from a patient diagnosed with glioma and used glioma cell lines to validate the performance of the GSScore in predicting chemotherapeutic responses. Differentially expressed genes (DEGs) between GSCs with high and low GSScores were used to cluster lower-grade glioma (LGG) samples into three stemness subtypes. Differences in clinicopathological characteristics, including survival, copy number variations, mutations, tumor microenvironment, and immune and chemotherapeutic responses, among the three LGG stemness-associated subtypes were identified. Using machine learning methods, we further identified genes as subtype predictors and validated their performance using the CGGA datasets. In the current study, we identified a GSScore that correlated with LGG chemotherapeutic response. Through the score, we also identified a novel classification of the LGG subtype and associated subtype predictors, which might facilitate the development of precision therapy.

Keywords

• Lower grade glioma
• Stemness
• Subtype
• Machine learning
• Prognosis