Single-cell RNA-sequencing data clustering using variational graph attention auto-encoder with self-supervised...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) enables cellular heterogeneity studies by analyzing gene expression at the individual cell level.
• scRNA-seq data analysis, particularly clustering, is crucial for identifying cell types and differentially expressed genes.
• Existing clustering methods struggle with scRNA-seq data's high dimensionality, sparsity, and dropout events, leading to suboptimal performance.

Purpose of the Study:

• To develop a novel deep learning framework for improved scRNA-seq data clustering.
• To address the limitations of current clustering algorithms when applied to high-dimensional, sparse scRNA-seq datasets.

Main Methods:

• A variational graph attention auto-encoder (VGAAE) deep learning framework was constructed for scRNA-seq data clustering.
• A multi-head attention mechanism was integrated to learn robust low-dimensional data representations.
• Self-supervised learning and Jaccard index were employed for cluster refinement and automatic determination of the number of clusters.

Main Results:

• The proposed VGAAE framework demonstrated superior performance in scRNA-seq data clustering compared to existing state-of-the-art methods.
• Experiments on diverse datasets validated the effectiveness of VGAAE in handling high dimensions, sparsity, and dropout events inherent in scRNA-seq data.
• The multi-head attention mechanism and self-supervised learning contributed to more accurate and robust clustering results.

Conclusions:

• The VGAAE framework offers a promising solution for overcoming the challenges in scRNA-seq data clustering.
• This approach enhances the identification of cell types and gene expression patterns, advancing the study of cell heterogeneity.
• VGAAE represents a significant advancement in applying deep learning for the analysis of single-cell genomics data.