PartIES: a disease subtyping framework with Partition-level Integration using diffusion-Enhanced Similarities from...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Multi-omics data integration is crucial for identifying complex disease subtypes.
• Existing methods often fail to preserve data-type-specific clustering structures and are sensitive to noisy similarity measures.

Purpose of the Study:

• To develop a novel method, PartIES (Partition-level Integration using diffusion-Enhanced Similarities), for robust disease subtyping using multi-omics data.
• To improve upon existing similarity-based subtyping methods by preserving data-type-specific information and reducing noise.

Main Methods:

• PartIES employs a diffusion step to denoise individual omics data similarity matrices.
• It extracts partition information from diffusion-enhanced similarities and integrates them iteratively via weighted averaging.

Main Results:

• Simulation studies demonstrated that the diffusion step significantly enhances clustering accuracy.
• PartIES outperformed competing methods, especially when omics data exhibited divergent clustering structures.
• Applied to The Cancer Genome Atlas data, PartIES identified novel subtypes in bladder, liver, and thyroid cancers significantly associated with patient survival.

Conclusions:

• PartIES offers a robust and effective approach for disease subtyping using multi-omics data.
• The identified cancer subtypes show distinct biological characteristics, including interactions with known cancer genes and differential pathway activity.
• The method provides valuable insights into cancer heterogeneity and patient outcomes.