HyGAnno: hybrid graph neural network-based cell type annotation for single-cell ATAC sequencing data

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Accurate cell type annotation is essential for single-cell omics data analysis.
• Current methods struggle with scATAC-seq data due to sparsity and cross-dataset inconsistencies.
• Existing approaches often rely solely on gene expression or activity, limiting their scope.

Purpose of the Study:

• To develop an automated method for cell type annotation in scATAC-seq data.
• To enable transfer of cell type labels from scRNA-seq references to scATAC-seq targets.
• To improve the reliability and interpretability of single-cell data analysis.

Main Methods:

• A novel semi-supervised method, HyGAnno, was developed.
• Utilizes a parallel graph neural network to transfer cell type information.
• Leverages genome-wide chromatin accessibility peaks and reconstructs a reference-target cell graph.

Main Results:

• HyGAnno demonstrates precise cell annotation across diverse datasets.
• The method generates interpretable cell embeddings.
• It shows robustness to noisy reference data and adaptability to tumor samples.

Conclusions:

• HyGAnno offers a robust solution for scATAC-seq cell annotation.
• The approach enhances the utility of scATAC-seq data by enabling accurate cell type identification.
• This method has broad applicability in single-cell genomics research, including cancer studies.