scGCC: Graph Contrastive Clustering With Neighborhood Augmentations for scRNA-Seq Data Analysis

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) is crucial for analyzing cellular heterogeneity.
• Cell clustering in scRNA-seq data is vital for identifying cell types and subtypes.
• scRNA-seq data presents challenges like high dimensionality, sparsity, and batch effects for clustering.

Purpose of the Study:

• To propose scGCC, a novel graph self-supervised contrastive learning model for scRNA-seq data clustering.
• To address computational challenges in scRNA-seq data analysis, improving cell type identification.
• To enhance the accuracy and robustness of cell clustering in scRNA-seq datasets.

Main Methods:

• Developed scGCC, a graph self-supervised contrastive learning model with representation and clustering modules.
• Employed Graph Attention Networks (GAT) for cell representation learning and feature extraction.
• Implemented five data augmentation methods to increase data diversity and reduce overfitting.

Main Results:

• scGCC learns low-dimensional denoised embeddings beneficial for clustering.
• Achieved extraordinary accuracy and robustness across 14 real-world scRNA-seq datasets.
• Demonstrated biological effectiveness through downstream tasks like batch effect removal and trajectory inference.

Conclusions:

• scGCC effectively addresses challenges in scRNA-seq data clustering.
• The model improves cell type identification and discovery of novel subtypes.
• scGCC offers a robust and accurate approach for scRNA-seq data analysis.