Differentiable graph clustering with structural grouping for single-cell RNA-seq data.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) analysis is crucial for understanding cellular heterogeneity.
• Deep graph clustering methods model cell relationships but often overlook inherent cluster structures.
• Bridging cell features and structural information in graph clustering remains a challenge.

Purpose of the Study:

• To develop a novel deep graph clustering method for scRNA-seq data.
• To integrate graph cluster information into deep clustering models.
• To enhance the accuracy of cell subpopulation identification.

Main Methods:

• Proposes Differentiable Graph Clustering with Structural Grouping (DGCSG).
• Employs an interactive module for layer-by-layer transfer of node representations between autoencoders (AE) and graph attention autoencoders (GATE).
• Introduces a differentiable clustering mechanism using spectral relaxation of K-way normalized cuts.
• Utilizes decoupled self-supervised optimization for representation learning.

Main Results:

• DGCSG effectively incorporates graph cluster information into deep graph clustering.
• The differentiable clustering mechanism learns clustering-friendly representations.
• DGCSG demonstrates superior performance on 14 scRNA-seq benchmarks compared to state-of-the-art methods.

Conclusions:

• DGCSG offers an advanced approach for scRNA-seq data analysis.
• The method enhances the identification of cell subpopulations by leveraging graph structure.
• DGCSG represents a significant improvement in deep graph clustering for single-cell genomics.