Clustering single-cell RNA sequencing data via iterative smoothing and self-supervised discriminative embedding.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) enables gene expression analysis at the individual cell level.
• Clustering is crucial for cell type identification and interaction discovery in scRNA-seq data.
• Robust similarity metrics for scRNA-seq clustering remain a challenge due to data complexity and noise.

Purpose of the Study:

• To introduce scRISE, a novel deep clustering method for scRNA-seq data.
• To address the challenge of selecting appropriate similarity metrics in scRNA-seq clustering.
• To enhance data representation and clustering accuracy for scRNA-seq datasets.

Main Methods:

• scRISE employs an iterative smoothing module using graph autoencoders for data denoising and similarity refinement.
• A self-supervised discriminative embedding module with an adaptive similarity threshold is utilized for accurate sample partitioning.
• The method integrates cell structural features and enriches data information through iterative smoothing.

Main Results:

• scRISE demonstrated superior data representation and clustering quality across seventeen scRNA-seq datasets.
• Performance was validated against multiple state-of-the-art deep learning clustering methods.
• Application to the HNSCC dataset identified 62 informative genes with potential therapeutic and biomarker roles.

Conclusions:

• scRISE offers an effective deep clustering solution for scRNA-seq data analysis.
• The method enhances the identification of cell types and biological insights from complex single-cell data.
• Identified genes present novel therapeutic targets and biomarkers for HNSCC.