A unified graph model based on molecular data binning for disease subtyping.

Area of Science:

• Computational biology
• Bioinformatics
• Precision medicine

Background:

• Molecular disease subtype discovery from omics data is crucial for precision medicine.
• Skewed distributions and data variability in omics data present significant challenges.
• Existing kernel-based methods lack robustness due to limitations in handling extreme values and data variability.

Purpose of the Study:

• To propose a novel robust distance metric (ROMDEX) for constructing patient similarity graphs.
• To address the challenges of data variability and extreme values in omics data for disease subtyping.
• To improve the identification of molecular disease subtypes with distinct clinical outcomes, such as survival.

Main Methods:

• Development of a novel robust distance metric (ROMDEX).
• Construction of patient similarity graphs using ROMDEX from omics data (Gene Expression, DNA Methylation, MicroRNA).
• Validation on multiple TCGA cancer datasets and comparison with state-of-the-art methods (MRGC, PINS, SNF, Consensus Clustering, Icluster+).

Main Results:

• ROMDEX demonstrated superior performance in molecular disease subtype discovery across different omics data types.
• Achieved best P-values (e.g., 0.00181 for Gene Expression) in Kaplan-Meier survival analysis, indicating significant survival differences between identified subtypes.
• Outperformed existing methods in identifying clinically relevant disease subtypes on TCGA datasets.

Conclusions:

• The proposed ROMDEX metric effectively addresses data variability and extreme values in omics data.
• ROMDEX enables more robust and accurate molecular disease subtype discovery.
• This approach holds significant potential for advancing precision medicine through improved patient stratification.