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

Area of Science:

• Computational biology
• Pediatric oncology
• Genomics

Background:

• Medulloblastoma (MB) is a common pediatric brain tumor with significant molecular diversity.
• Accurate classification of MB subgroups is crucial for tailoring treatments and predicting outcomes.
• Existing classification methods may not fully capture the molecular heterogeneity of MB.

Purpose of the Study:

• To develop and evaluate machine learning (ML) models for classifying medulloblastoma molecular subgroups.
• To assess the impact of feature selection on the performance of ML classifiers for MB subgroup identification.
• To identify optimal ML approaches for distinguishing between different medulloblastoma subtypes.

Main Methods:

• RNA sequencing data from 70 pediatric medulloblastoma samples were analyzed.
• Five ML classifiers (KNN, DT, SVM, RF, NB) were employed to predict MB subgroups.
• Feature selection was performed to identify optimal gene subsets (750, 75, 25 genes) for classification.

Main Results:

• Reduced gene sets significantly improved classification and clustering performance compared to the full gene set.
• Random Forest (RF), K-nearest Neighbors (KNN), and Support Vector Machine (SVM) classifiers demonstrated superior performance.
• High classification accuracies (>90%) were achieved, particularly for medulloblastoma groups 3 and 4, using optimized feature sets.

Conclusions:

• Machine learning algorithms are effective tools for classifying medulloblastoma molecular subgroups based on gene expression.
• Feature selection is a critical step for enhancing the accuracy and efficiency of ML-based MB classification.
• These findings support the development of advanced, personalized treatment strategies for medulloblastoma patients.