cytoKernel: Robust kernel embeddings for assessing differential expression of single cell data.

Area of Science:

• Single-cell omics
• Computational biology
• Biostatistics

Background:

• Existing differential expression methods often focus on aggregate measurements, missing complex patterns in single-cell data.
• High-throughput sequencing and cytometry generate complex, multimodal data distributions.
• Detecting subtle, non-global expression variations is crucial for understanding cell specification.

Purpose of the Study:

• To introduce cytoKernel, a robust kernel-based score test for differential expression analysis of single-cell data.
• To enable detection of both aggregate and subtle differential expression patterns.
• To provide a method that utilizes the full probability distribution of single-cell data.

Main Methods:

• Kernel embeddings are used to compute pairwise divergence between probability distributions of subjects.
• A kernel-based score test is applied to assess differential expression.
• The method is benchmarked on simulated and real single-cell RNA sequencing and mass cytometry datasets.

Main Results:

• cytoKernel effectively controls the False Discovery Rate (FDR).
• The method demonstrates superior performance compared to existing approaches in identifying differential patterns.
• cytoKernel successfully detects subtle variations often overlooked by traditional methods.

Conclusions:

• cytoKernel offers a powerful new approach for differential expression analysis in single-cell genomics and cytometry.
• The method enhances the ability to identify complex biological variations from high-dimensional single-cell data.
• The cytoKernel R package is available for broader scientific application.