Machine Learning–Driven Identification of Molecular Subgroups in Medulloblastoma via Gene Expression Profiling

Abstract

Aims

Medulloblastoma (MB) is the most prevalent malignant brain tumour in children, characterised by substantial molecular heterogeneity across its subgroups. Accurate classification is pivotal for personalised treatment strategies and prognostic assessments. In this study, we aimed to build machine learning models to classify MB subgroups.

Materials and Methods

This study utilised machine learning (ML) techniques to analyse RNA sequencing data from 70 paediatric MB samples. Five classifiers—K-nearest neighbors (KNN), decision tree (DT), support vector machine (SVM), random forest (RF), and naive Bayes (NB)—were used to predict molecular subgroups based on gene expression profiles. Feature selection identified gene subsets of varying sizes (750, 75, and 25 genes) to optimise classification accuracy.

Results

Initial analyses with the complete gene set lacked discriminative power. However, reduced feature sets significantly enhanced clustering and classification performance, particularly for group 3 and group 4 subgroups. The RF, KNN, and SVM classifiers consistently outperformed the DT and NB classifiers, achieving classification accuracies exceeding 90% in many scenarios, especially in group 3 and group 4 subgroups.

Conclusion

This study highlights the efficacy of ML algorithms in classifying MB subgroups using gene expression data. The integration of feature selection techniques substantially improves model performance, paving the way for enhanced personalised approaches in MB management.