A copula based topology preserving graph convolution network for clustering of single-cell RNA-seq data.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) analysis requires homogeneous cell grouping for accurate annotation.
• Challenges include low RNA input, limited reads, cell-cycle variations, and technical noise impacting feature selection for clustering.

Purpose of the Study:

• Introduce sc-CGconv, a novel unsupervised approach for robust feature extraction and cell clustering in scRNA-seq data.
• Address limitations of existing methods in handling noise and variability for improved cell population identification.

Main Methods:

• sc-CGconv employs copula correlation (Ccor) to formulate cell-cell relationships, creating a graph structure.
• A graph convolution network (GCN) learns representations from this graph for unsupervised clustering.
• The method integrates feature extraction and clustering in a stepwise manner.

Main Results:

• sc-CGconv effectively identifies homogeneous cell clusters using substantially smaller sample sizes.
• It models expression co-variability across numerous genes, surpassing current state-of-the-art feature selection methods.
• The approach preserves cell-to-cell variability and provides topology-preserving low-dimensional embeddings.

Conclusions:

• sc-CGconv offers a robust and efficient method for scRNA-seq data analysis and cell clustering.
• Its ability to handle noise and variability makes it a valuable tool for biological discovery.