Heterogeneity-preserving discriminative feature selection for disease-specific subtype discovery.

Area of Science:

• Computational Biology
• Genomics
• Biostatistics

Background:

• Disease heterogeneity complicates understanding progression and personalized therapies.
• High-dimensional omics data offer opportunities but pose challenges for subtype discovery.
• Existing feature selection methods often miss features crucial for revealing novel subtypes.

Purpose of the Study:

• To develop a statistical methodology for identifying disease subtypes that preserves sample heterogeneity.
• To identify a minimal set of features that improve subtype clustering quality.
• To overcome limitations of current methods in discovering new disease subtypes.

Main Methods:

• Developed Preserving Heterogeneity (PHet), a statistical methodology.
• Employed iterative subsampling and differential analysis of interquartile range.
• Integrated Fisher's method for feature selection to enhance clustering.

Main Results:

• PHet successfully maintains sample heterogeneity while distinguishing known disease/cell states.
• The method identified a small set of features that significantly enhance subtype clustering.
• PHet demonstrated superior performance compared to previous differential expression and outlier-based methods.

Conclusions:

• PHet offers a novel approach to subtype discovery by preserving heterogeneity.
• The methodology has the potential to advance the understanding of disease mechanisms.
• This approach can aid in identifying new subtypes and informing personalized therapies.