Dual-GCN-based deep clustering with triplet contrast for ScRNA-seq data analysis

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Single-cell RNA sequencing (ScRNA-seq) provides high-resolution gene expression data.
• Clustering and dimensionality reduction are essential for ScRNA-seq data analysis.
• Existing graph-based deep clustering methods may overlook node distribution and suffer from over-smoothing, impacting embedding quality.

Purpose of the Study:

• To propose a novel dual-GCN-based deep clustering method with Triplet contrast (scDGDC) for ScRNA-seq data.
• To enhance the capture of topological and distribution information in ScRNA-seq data analysis.
• To improve the performance of dimensionality reduction and clustering tasks.

Main Methods:

• Developed a dual-GCN-based encoder to capture comprehensive topological information.
• Incorporated Triplet contrast to mitigate the over-smoothing issue common in graph convolutional networks (GCNs).
• Evaluated scDGDC on eight real-world ScRNA-seq datasets.

Main Results:

• scDGDC demonstrated excellent performance in both dimensionality reduction and clustering tasks.
• The dual-GCN encoder effectively captured richer topological information.
• Triplet contrast successfully reduced GCN over-smoothing, leading to improved embedding representations.
• The method showed high robustness to parameter variations.

Conclusions:

• scDGDC offers an effective approach for analyzing ScRNA-seq data by integrating topological and distribution information.
• The proposed method addresses limitations of existing graph-based deep clustering algorithms.
• scDGDC provides a robust and high-performing solution for ScRNA-seq dimensionality reduction and clustering.