GCLSC: Single-cell clustering model based on graph contrastive learning.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) reveals cellular heterogeneity.
• Cell clustering is vital for identifying cell types and subtypes in scRNA-seq data.
• Challenges in scRNA-seq data include high dimensionality, sparsity, and technical artifacts.

Purpose of the Study:

• To develop a novel graph contrastive learning model for robust single-cell clustering.
• To address the challenges posed by scRNA-seq data characteristics for accurate cell clustering.
• To provide a reliable computational tool for cell population profiling.

Main Methods:

• Proposed GCLSC (Graph Contrastive Learning for Single-Cell Clustering) model.
• Integrated Graph Transformer and Graph Attention Network (GAT) to model cellular interactions and dependencies.
• Employed four data augmentation strategies to enhance data diversity and prevent overfitting.

Main Results:

• GCLSC achieved superior clustering accuracy across nine real-world scRNA-seq datasets.
• Demonstrated effectiveness in identifying novel cell subtypes and annotating known cell types.
• Validated as a reliable tool for cell population profiling.

Conclusions:

• GCLSC effectively combines GAT, Transformer, and contrastive learning for robust single-cell analysis.
• The model offers significant improvements in cell clustering accuracy for scRNA-seq data.
• Accurate clustering supports critical downstream analyses in single-cell research.