Kernel-based testing for single-cell differential analysis.

Area of Science:

• Computational biology
• Genomics
• Epigenetics

Background:

• Single-cell technologies provide high-resolution molecular data but comparing cell populations remains challenging.
• Existing methods may struggle to capture complex, non-linear differences in cell-wise distributions.
• Understanding cell heterogeneity is crucial for fields like developmental biology and disease research.

Purpose of the Study:

• To introduce a novel kernel-testing framework for non-linear, cell-wise distribution comparison.
• To enable feature-wise and global comparisons of transcriptomic and epigenomic data.
• To identify subtle cell population variations and transitions in cell states.

Main Methods:

• Development of a kernel-testing framework for comparing non-linear distributions.
• Application to gene expression and epigenomic modification data.
• Utilizing a classifier based on embedding variability for state transition identification.

Main Results:

• The kernel-testing framework successfully performs feature-wise and global transcriptome/epigenome comparisons.
• The method reveals significant cell population heterogeneities and subtle variations.
• Analysis of single-cell ChIP-Seq data identified untreated breast cancer cells with a persister-like epigenomic profile.

Conclusions:

• Kernel testing provides an effective approach for uncovering subtle population variations missed by traditional single-cell analyses.
• The framework advances the comparison of complex molecular feature distributions in single cells.
• This method has potential applications in identifying specific cell subpopulations and understanding disease states.