scGANSL: Graph Attention Network with Subspace Learning for scRNA-seq Data Clustering.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) enables detailed analysis of cellular heterogeneity.
• Cell clustering is vital for identifying cell types and subpopulations in scRNA-seq data.
• Existing methods often use single views, limiting interpretation and struggling with high dimensionality and noise.

Purpose of the Study:

• To develop a novel, robust clustering method for scRNA-seq data.
• To address the challenges of high dimensionality and noise in scRNA-seq analysis.
• To improve the accuracy of cell type identification and subpopulation discovery.

Main Methods:

• Introduced scGANSL, a graph attention network with subspace learning for scRNA-seq clustering.
• Constructed two views using highly variable genes (HVGs) screening and principal component analysis (PCA).
• Integrated a multi-view shared graph autoencoder, zero-inflated negative binomial (ZINB) model, and self-supervised graph attention autoencoder.
• Employed a local learning and self-expression strategy to preserve data structure.

Main Results:

• scGANSL demonstrated superior performance compared to existing state-of-the-art methods.
• The method effectively handled high-dimensional and noisy scRNA-seq data.
• Experimental results validated the model's ability to accurately cluster cells across diverse datasets.

Conclusions:

• scGANSL offers a significant advancement in scRNA-seq data clustering.
• The multi-view approach and integrated models enhance the interpretation of cellular diversity.
• This method provides a more comprehensive understanding of biological systems through precise cell clustering.