Detecting molecular subtypes from multi-omics datasets using SUMO.

Area of Science:

• Oncology
• Bioinformatics
• Genomics

Background:

• Accurate molecular subtyping is crucial for understanding cancer heterogeneity and improving patient outcomes.
• Integrating multi-omics data presents challenges due to data complexity and missing values.

Purpose of the Study:

• To present a data integration framework for molecular subtyping using non-negative matrix factorization (NMF) of patient-similarity networks.
• To evaluate the framework's ability to handle missing data and detect clinically relevant cancer subtypes.

Main Methods:

• Utilized non-negative matrix factorization (NMF) on patient-similarity networks.
• Integrated continuous multi-omics datasets for molecular subtyping.
• Applied the framework to lower-grade glioma data.

Main Results:

• The framework effectively integrates multi-omics data and handles missing values without imputation.
• Consistently identified subtypes with differential prognosis and clinical associations across various cancers.
• Discovered a distinct lower-grade glioma subtype with worse prognosis, characterized by hypomethylation and specific chromosomal alterations (chr7 gain, chr10 loss).

Conclusions:

• The proposed data integration framework is a powerful tool for molecular subtyping and cancer research.
• Identified a novel, prognostically relevant subtype of lower-grade gliomas with potential diagnostic implications.
• The findings support the integration of advanced molecular markers into future cancer diagnostic guidelines.