scGAC: a graph attentional architecture for clustering single-cell RNA-seq data.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) enables cellular-level biological research.
• Clustering scRNA-seq data into cell subpopulations is crucial but challenging due to data complexity (high variability, sparsity, dimensionality).
• Existing clustering methods often fail to fully utilize latent relationships among cells, leading to suboptimal results.

Purpose of the Study:

• To propose a novel unsupervised clustering method, scGAC (single-cell Graph Attentional Clustering), for scRNA-seq data.
• To address the limitations of existing methods in capturing latent cellular relationships.
• To improve the accuracy and performance of single-cell data clustering.

Main Methods:

• scGAC constructs and refines a cell graph using network denoising.
• It employs a graph attentional autoencoder to learn clustering-friendly cell representations by propagating information with weighted importance.
• A self-optimizing approach is used to determine the final cell clusters.

Main Results:

• scGAC demonstrates excellent performance across 16 real-world scRNA-seq datasets.
• The method effectively captures latent relationships among cells.
• scGAC outperforms existing state-of-the-art single-cell clustering techniques.

Conclusions:

• scGAC offers a powerful new approach for unsupervised clustering of scRNA-seq data.
• The method's ability to capture complex cellular relationships leads to improved clustering accuracy.
• scGAC provides a valuable tool for advancing biological research using scRNA-seq data.