scMGCN: A Multi-View Graph Convolutional Network for Cell Type Identification in scRNA-seq Data.

Area of Science:

• Computational Biology
• Genomics
• Artificial Intelligence in Medicine

Background:

• Single-cell RNA sequencing (scRNA-seq) provides deep insights into cellular diversity and molecular interactions.
• Accurate cell type identification from scRNA-seq data is crucial but challenging due to data heterogeneity and batch effects.
• Existing cell identification methods often lack stability and interpretability.

Purpose of the Study:

• To develop a robust graph artificial intelligence model for stable and accurate cell type identification from scRNA-seq data.
• To address limitations of current methods in handling cellular heterogeneity and batch effects.
• To improve the interpretability and performance of cell type prediction.

Main Methods:

• Developed a multi-view graph convolutional network model (scMGCN) integrating multiple graph structures from scRNA-seq data.
• Employed graph convolutional networks with attention mechanisms to learn cell embeddings and predict cell labels.
• Utilized multi-view learning and diverse graph construction techniques to capture comprehensive cellular information.

Main Results:

• scMGCN demonstrated superior stability, accuracy, and robustness to batch effects compared to state-of-the-art methods.
• The model effectively extracted shared, high-order information from cells using graph convolutional networks and attention.
• Evaluated scMGCN's performance across single-dataset, cross-species, and cross-platform experiments, confirming its effectiveness.

Conclusions:

• scMGCN offers a powerful and reliable approach for cell type identification in scRNA-seq data.
• The multi-view graph learning framework enhances the ability to analyze complex cellular ecosystems.
• This model advances the field of computational biology and biomedical research by improving data analysis capabilities.