SSRE

SSRE generates sparse subspace representations and enhances cell-to-cell similarity to improve cell type identification from single-cell RNA sequencing (scRNA-seq) data.

Key Features:

• Sparse subspace representation: Produces sparse representations of cell-to-cell similarities under a subspace assumption, retaining the most relevant neighbors for each cell.
• Similarity enhancement: Refines similarity measurements between cells through a similarity enhancement procedure.
• Three classical pairwise similarities: Integrates and refines three classical pairwise similarity measures to complement the sparse representation.
• Gene selection and enhancement strategy: Applies gene selection and an enhancement strategy to improve similarity estimation and downstream clustering.
• Unsupervised clustering: Frames cell type identification as an unsupervised clustering problem using the enhanced similarity matrix.
• Visualization of scRNA-seq data: Provides functionality for visualizing cell populations from scRNA-seq datasets.
• Differentially expressed gene identification: Identifies differentially expressed genes among inferred cell clusters.
• Benchmarking: Demonstrated performance on ten real-world scRNA-seq datasets and five simulated datasets, outperforming several state-of-the-art single-cell clustering methods as reported.

Scientific Applications:

• Cell type identification: Unsupervised identification of distinct cell types from scRNA-seq data.
• Discrimination of complex cell populations: Distinguishing different cell types within complex scRNA-seq datasets.
• scRNA-seq data visualization: Visual interpretation of cell populations and similarity structure in scRNA-seq studies.
• Differential expression analysis: Detection of differentially expressed genes across inferred clusters.

Methodology:

SSRE models cell relationships under a subspace assumption to generate sparse cell-to-cell similarity representations, refines similarities by integrating three classical pairwise similarity measures and a gene selection/enhancement strategy, and applies unsupervised clustering.