sc-CGconv

sc-CGconv generates topology-preserving low-dimensional embeddings for clustering single-cell RNA sequencing (scRNA-seq) data by combining copula correlation (Ccor) and graph convolution networks (GCN).

Key Features:

• Copula Correlation for Cell-Cell Relationships: sc-CGconv uses copula correlation (Ccor) to model and aggregate cell-cell relationships and capture expression co-variability across large gene sets.
• Graph Convolution Network Integration: A graph convolution network (GCN) learns from the cell-cell graph constructed using Ccor to produce low-dimensional embeddings.
• Topology-Preserving Embedding: The learned embeddings preserve the topological structure of the original high-dimensional scRNA-seq data.
• Robustness to Technical Noise: Robust unsupervised feature extraction using Ccor mitigates technical noise inherent in scRNA-seq and improves gene selection/extraction.
• Efficiency with Small Sample Sizes: The method identifies homogeneous clusters from datasets with limited cell numbers.

Scientific Applications:

• Single-Cell Clustering: sc-CGconv performs unsupervised clustering of scRNA-seq data to recover biologically meaningful cell groups while handling variability from cell cycle, morphology, and reagent concentrations.
• Gene Expression Co-Variability Modeling: By modeling co-variability of gene expression across cells, sc-CGconv facilitates analysis of cellular heterogeneity.

Methodology:

Compute copula correlation to extract cell-cell relationships, construct a graph from these correlations, process the graph with a graph convolution network (GCN) to learn a low-dimensional representation, and use the learned representations for clustering.