Self-Supervised Graph Representation Learning for Single-Cell Classification.

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Accurate cell type identification from single-cell RNA sequencing (scRNA-seq) data is crucial for biological research.
• Traditional methods are time-consuming, necessitating advanced computational approaches.
• Existing computational methods struggle to fully leverage unlabeled scRNA-seq data, limiting classification accuracy and generalizability.

Purpose of the Study:

• To propose a novel self-supervised graph representation learning framework, scSSGC, for enhanced single-cell classification.
• To address the challenge of limited labeled data in scRNA-seq analysis.
• To improve the utilization of gene expression information for robust cell identification.

Main Methods:

• Developed scSSGC, a self-supervised graph representation learning framework.
• Employed multiple K-means clustering tasks on unlabeled data for model pre-training.
• Introduced a locally augmented graph neural network to capture cell interactions and enhance information aggregation.

Main Results:

• scSSGC demonstrated superior performance compared to existing state-of-the-art cell classification methods in benchmark experiments.
• The framework achieved stable performance on cross-dataset evaluations, indicating strong generalization ability.
• Effective utilization of unlabeled gene expression data was achieved, mitigating limitations of sparse labeled datasets.

Conclusions:

• scSSGC offers a powerful new approach for accurate and generalizable single-cell classification.
• The self-supervised learning strategy effectively overcomes data limitations in scRNA-seq analysis.
• This framework advances computational methods for understanding cellular differentiation and disease mechanisms.