A robust and scalable graph neural network for accurate single-cell classification.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) generates high-resolution data on cellular heterogeneity.
• Accurate cell type identification is crucial for scRNA-seq data analysis.
• Traditional methods are time-consuming and subjective, limiting scalability for large datasets.

Purpose of the Study:

• To develop a robust and scalable graph neural network (GNN)-based method for accurate single-cell classification.
• To enable efficient cell type identification in large-scale scRNA-seq datasets by transferring labels from annotated data.

Main Methods:

• GraphCS constructs a graph connecting similar cells across labeled and unlabeled scRNA-seq datasets.
• Pre-calculates diffused information using approximate Generalized PageRank for sublinear complexity.
• Employs a GNN for efficient information propagation and cell classification.

Main Results:

• GraphCS demonstrates superior performance on simulated, cross-platform, cross-species, and cross-omics scRNA-seq datasets.
• Achieves high speed and scalability, classifying 1 million cells within 50 minutes.
• Outperforms existing methods in accuracy and efficiency for large-scale single-cell data.

Conclusions:

• GraphCS provides a powerful and efficient solution for single-cell classification in the era of big data.
• The method overcomes the scalability limitations of traditional GNNs for analyzing millions of cells.
• Enables accurate and rapid cell type identification, advancing scRNA-seq data analysis.