Comprehensive multi-omics and machine learning framework for glioma subtyping and precision therapeutics

Area of Science:

• Neuro-oncology
• Computational biology
• Cancer immunology

Background:

• Glioma is a highly aggressive brain tumor with complex molecular and immunological characteristics.
• Understanding these features is crucial for developing effective therapeutic strategies.

Purpose of the Study:

• To integrate multi-omics data for classifying glioma subtypes.
• To develop a robust prognostic signature for glioma.
• To identify potential therapeutic strategies based on molecular subtypes.

Main Methods:

• Utilized the MOVICS framework to derive three integrative molecular subtypes (CS1, CS2, CS3) from multi-omics data of TCGA diffuse-glioma patients.
• Benchmarked ten machine-learning algorithms in MIME, selecting Lasso + SuperPC to develop an eight-gene GloMICS prognostic score.
• Performed TME deconvolution, immune checkpoint profiling, and TIDE analysis.

Main Results:

• Identified three distinct glioma subtypes (CS1: astrocyte-like, CS2: mesenchymal, CS3: metabolic) with differential biology and prognoses.
• The eight-gene GloMICS score demonstrated superior prognostic performance across multiple cohorts compared to 95 existing models.
• High-risk GloMICS tumors exhibit immunosuppressive microenvironments, suggesting potential benefits from immunotherapy or targeted agents like dabrafenib and irinotecan.

Conclusions:

• Established robust glioma subtypes and a transferable prognostic signature (GloMICS) for improved patient stratification.
• The findings provide a blueprint for biomarker-guided therapy in glioma.
• Further validation and prospective trials are warranted to translate these findings into clinical practice.