Characterization of immune microenvironment associated with medulloblastoma metastasis based on explainable machine learning.

Importance: Medulloblastoma (MB) is the most common malignant brain tumor in children, with metastasis being the primary cause of recurrence and mortality. The tumor microenvironment (TME) plays a critical role in driving metastasis; however, the mechanisms underlying TME alterations in MB metastasis remain poorly understood. Objective: To develop and validate machine learning (ML) models for predicting patient outcomes in MB and to investigate the role of TME components, particularly immune cells and immunoregulatory molecules, in metastasis. Methods: ML models were constructed and validated to predict prognosis and metastasis in MB patients. Eight algorithms were evaluated, and the optimal model was selected. Lasso regression was employed for feature selection, and SHapley Additive exPlanations values were used to interpret the contribution of individual features to model predictions. Immune cell infiltration in tumor tissues was quantified using the microenvironment cell populations-counter method, and immunohistochemistry was applied to analyze the expression and distribution of specific proteins in tumor tissues. Results: The ML models identified metastasis as the strongest predictor of poor prognosis in MB patients, with significantly worse survival outcomes observed in metastatic cases. High infiltration of CD8+ T cells and cytotoxic T lymphocytes (CTLs), along with elevated expression of the Interpretation: This study identifies metastasis as the key prognostic factor in MB and reveals the pivotal roles of CD8+ T cells, CTLs, and TGF-β1 within the TME in promoting metastasis and poor outcomes. These findings provide a foundation for developing future therapeutic strategies targeting the TME to improve MB patient outcomes.