Deep single-cell RNA-seq data clustering with graph prototypical contrastive learning.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) is crucial for studying cellular heterogeneity.
• Accurate cell type identification is vital for downstream scRNA-seq analysis.
• Challenges like data sparsity (dropout phenomena) and noise impede robust clustering.

Purpose of the Study:

• To develop a robust method for cell type identification in scRNA-seq data.
• To address limitations of existing methods in handling noisy data and leveraging relational information.
• To propose a graph-based prototypical contrastive learning approach for improved cell representation.

Main Methods:

• A graph-based prototypical contrastive learning method (scGPCL) is proposed.
• Cell representations are encoded using Graph Neural Networks on a cell-gene graph.
• Prototypical contrastive learning is employed to refine cell representations by contrasting dissimilar and similar cell pairs.

Main Results:

• scGPCL effectively leverages relational information inherent in scRNA-seq data.
• The method demonstrates improved cell representation learning compared to existing approaches.
• Extensive experiments on simulated and real data confirm scGPCL's effectiveness and efficiency.

Conclusions:

• scGPCL offers a powerful new approach for cell type identification in scRNA-seq data.
• The method successfully mitigates challenges posed by data sparsity and noise.
• The proposed graph-based prototypical contrastive learning framework enhances the analysis of single-cell data.