scQcut

scQcut performs parameter-free graph-based clustering of single-cell RNA sequencing (scRNAseq) data to identify cell subtypes from gene expression profiles.

Key Features:

• Parameter-Free Operation: Operates without manual parameter tuning, removing the need for user-defined clustering parameters.
• Graph-Based Clustering: Constructs a k-nearest-neighbor (KNN) network to represent relationships between cells based on expression profiles.
• Optimal K Estimation: Estimates the optimal value of k using a topology-based criterion to guide KNN graph construction.
• Modularity-Based Community Discovery: Applies an efficient modularity-based community discovery algorithm (modularity optimization) to predict cell clusters.
• Performance and Accuracy: Extensive testing on real and synthetic datasets demonstrated improved clustering accuracy relative to several state-of-the-art methods and enhanced detection of rare cell types.

Scientific Applications:

• Unsupervised Identification of Cell Subtypes: Enables discovery of novel cell subtypes from scRNAseq data without prior labels.
• Analysis of Complex Biological Systems: Handles sparse data and complex population topology for studies of development, disease mechanisms, and cellular heterogeneity at single-cell resolution.

Methodology:

Constructs a k-nearest-neighbor (KNN) graph, estimates optimal k via a topology-based criterion, and identifies clusters using modularity-based community discovery (modularity optimization).