Ensemble Machine Learning Approaches Predict Survival in Lower-Grade Glioma Based on Glycosphingolipid Gene Expression and Metabolic Modelling

Abstract

Glycosphingolipids (GSLs) are essential components of biological membranes with important roles in cell signalling. Disrupted GSL metabolism is associated with malignancy across a range of cancers, with different GSLs implicated in distinct tumours. GSLs have potential mechanistic roles in cancer; however, their functions in Lower Grade Gliomas (LGGs) remain poorly understood. We present ensemble machine learning approaches using transcriptomic data from LGG, combined with GSL-specific metabolic simulations, to predict survival outcomes. The ensemble approach demonstrates effective risk stratification for LGG patients based on GSL gene expression. Pathway analysis of model-derived risk groups highlighted potential association of GSLs with cell motility, division and Wnt signalling in LGG pathology. Given the strong performance of machine learning approaches to predict outcomes and that GSLs are shed into the tumour microenvironment, GSL-based diagnostics and prognostics may prove clinically beneficial. A Python package enabling GSL-specific metabolic modelling and risk prediction from RNA-seq data is provided.