RCSL

RCSL: Clustering algorithm for single-cell RNA-seq data

RCSL implements a clustering algorithm that integrates global and local similarity measures to identify heterogeneous cell types in single-cell RNA-seq (scRNA-seq) datasets from complex tissues.

Key Features:

• Novel Clustering Algorithm: Integrates local and global similarities to improve identification of heterogeneous cell types.
• Global Similarity Measurement: Computes global similarity using Spearman's rank correlations between cell expression vectors.
• Local Similarity Learning: Learns adaptive neighbor representations to model local similarities among cells.
• Linear Combination of Similarities: Calculates overall cell–cell similarity as a linear combination of global and local similarities.
• Automatic Estimation of Cell Types: Estimates the number of cell types by constructing a block-diagonal matrix that minimizes distance to the original similarity matrix, where each submatrix corresponds to a connected component in the similarity graph.

Scientific Applications:

• Single-Cell Transcriptomics: Identifies distinct and subtly different cell types in scRNA-seq data derived from complex tissues.

Methodology:

RCSL calculates global similarity using Spearman's rank correlations of gene expression vectors, learns local similarities through adaptive neighbor representation, combines these measures into a unified similarity metric, and constructs a block-diagonal matrix to identify connected components representing cell clusters while minimizing distance to the original similarity matrix.