Constructing metabolism-protein interaction relationship to identify glioma prognosis using deep learning

Area of Science:

• Oncology
• Bioinformatics
• Systems Biology

Background:

• Glioma exhibits significant heterogeneity, necessitating subtype classification for tailored treatments and prognosis.
• Metabolic-protein interactions (MPI) offer a promising avenue for understanding cancer complexity.
• The role of lipids and lactate in glioma subtyping remains underexplored.

Purpose of the Study:

• To develop a novel method for identifying glioma prognostic subtypes.
• To investigate the utility of metabolic-protein interactions (MPI) in classifying glioma.
• To explore the potential of lipids and lactate in defining glioma subtypes.

Main Methods:

• Constructed a metabolic-protein interaction relationship matrix (MPIRM) using a triple-layer network (Tri-MPN) integrated with mRNA expression data.
• Applied deep learning techniques to the MPIRM for identifying distinct glioma subtypes.
• Analyzed prognostic differences, immune infiltration, mutational signatures, and pathway signatures across identified subtypes.

Main Results:

• Identified statistically significant glioma prognostic subtypes (p-value < 2e-16, 95% CI).
• Demonstrated strong correlations between these subtypes and immune infiltration, mutational signatures, and pathway signatures.
• Validated the effectiveness of MPI network node interactions in characterizing glioma heterogeneity.

Conclusions:

• The proposed deep learning approach effectively identifies glioma prognostic subtypes based on MPI networks.
• MPI-based subtyping provides valuable insights into glioma heterogeneity, correlating with key biological features.
• This method holds potential for improving glioma classification and guiding personalized treatment strategies.