Single-cell Tree-based Model for Genomic-Disease Association.

Area of Science:

• Single-cell multi-omics
• Biomarker discovery
• Computational biology

Background:

• Single-cell multi-omics technologies offer high resolution for disease mapping and biomarker identification.
• Current methods often overlook combinatorial interactions of genomic features, complicating phenotype analysis.
• Identifying complex interactions is crucial for understanding clinical outcomes.

Purpose of the Study:

• To present scanCT, a tree-based framework for interpretable identification of genomic feature groups associated with disease phenotypes.
• To capture feature interactions and adjust for confounding factors in single-cell data.
• To apply scanCT to COVID-19 multi-omics data for biomarker discovery.

Main Methods:

• scanCT utilizes a tree-based framework with unbiased, model-based variable selection for data-driven split selection.
• The architecture captures feature interaction effects, enabling analysis of combinatorial biomarker patterns.
• Association modeling adjusts for confounding factors like age and sex.

Main Results:

• scanCT was applied to longitudinal single-cell multi-omics COVID-19 datasets.
• The framework identified phenotype-specific gene and protein markers.
• Interpretable synergistic marker combinations were revealed, explaining clinical phenotype variations.

Conclusions:

• scanCT provides a powerful and interpretable approach for analyzing single-cell multi-omics data.
• The framework effectively identifies complex biomarker interactions driving disease phenotypes.
• This method advances biomarker discovery for complex diseases like COVID-19.