SIMLR

SIMLR learns cell-to-cell similarity measures from single-cell RNA sequencing (scRNA-seq) data to enable improved dimension reduction, clustering, and visualization of heterogeneous cellular populations.

Key Features:

• Multi-kernel similarity-learning framework: Employs multi-kernel learning to derive a distance metric tailored to scRNA-seq data that captures intrinsic cell-to-cell similarities.
• Dimension reduction, clustering, and visualization: Applies the learned similarity metric to perform dimension reduction, clustering, and visualization of single-cell datasets.
• Improved subpopulation separation: Enables more accurate separation of known subpopulations within single-cell datasets compared to existing dimension reduction techniques.
• Benchmarking and performance: Demonstrated superior scalability and clustering performance through benchmarking against state-of-the-art methods on several public datasets.
• Validation on PBMC and GemCode data: Validated on high-throughput peripheral blood mononuclear cells (PBMC) datasets generated by GemCode, showing robustness for large-scale data.
• Interpretable visualizations: Produces more interpretable visualizations that aid exploration and interpretation of cellular heterogeneity.

Scientific Applications:

• scRNA-seq analysis: Analysis of single-cell RNA-seq datasets for identification and characterization of cellular populations.
• Cell differentiation studies: Investigating cellular heterogeneity during cell differentiation processes.
• Disease progression and oncology: Studying heterogeneity and progression in disease contexts, including oncology research.
• Immunology and PBMC studies: Analysis of immune cell populations such as peripheral blood mononuclear cells.

Methodology:

SIMLR learns a similarity measure that captures the intrinsic structure of scRNA-seq data via multi-kernel learning by combining multiple kernels to construct a comprehensive distance metric, and applies the learned metric to dimension reduction, clustering, and visualization.