Consensus clustering with missing labels (ccml): a consensus clustering tool for multi-omics integrative prediction...

Area of Science:

• Biomedical Data Science
• Computational Biology
• Bioinformatics

Background:

• Multi-omics data integration is crucial for understanding complex diseases.
• Existing consensus clustering methods struggle with missing data and varying sample coverages.
• Accurate patient stratification requires robust methods that handle real-world data limitations.

Purpose of the Study:

• To introduce a novel consensus clustering with missing labels (ccml) strategy.
• To enable effective multi-omics data integration despite unequal sample coverages and missing data.
• To provide a flexible R protocol for advanced biomedical data analysis.

Main Methods:

• Developed a two-step consensus clustering protocol (ccml) in R.
• Adjusted consensus weights by sample coverage to normalize for missing data.
• Applied ccml to 9-omics data in COPD (Karolinska COSMIC) and 24-omics data in asthma (U-BIOPRED).

Main Results:

• ccml successfully handled unequal missing labels across multiple omics datasets.
• Identified molecularly distinct subgroups in COPD and asthma cohorts.
• Demonstrated the utility of ccml for integrative subgrouping of complex diseases.

Conclusions:

• ccml is an effective tool for multi-omics integration, overcoming limitations of missing data.
• The method enhances downstream analysis for algorithms like Similarity Network Fusion.
• ccml provides a valuable resource for researchers analyzing complex human cohort data.