RGCN-BA: relational graph convolutional network with batch awareness for single-cell RNA sequencing clustering

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) provides cellular heterogeneity insights but requires accurate clustering and batch correction.
• Current methods often address cell clustering and batch effect correction separately, limiting their effectiveness across diverse scRNA-seq datasets.
• A unified approach is needed to simultaneously handle these critical analysis steps.

Purpose of the Study:

• To introduce Relational Graph Convolutional Network with Batch Awareness (RGCN-BA), a novel deep learning framework.
• To integrate cell clustering and batch effect correction into a single, versatile model for scRNA-seq data analysis.
• To demonstrate the superior performance of RGCN-BA compared to existing specialized methods.

Main Methods:

• RGCN-BA utilizes a relational graph convolutional network to model batch information as distinct edge types for multi-batch data.
• A batch correction layer ensures global data alignment within the unified framework.
• For single-batch data, the model operates with a single edge type, showcasing its adaptability.

Main Results:

• Experiments on both multi-batch and single-batch scRNA-seq datasets showed RGCN-BA's superior performance.
• RGCN-BA outperformed specialized methods for both cell clustering and batch effect correction.
• The unified approach demonstrated enhanced accuracy and robustness in scRNA-seq data analysis.

Conclusions:

• RGCN-BA offers a powerful, integrated solution for scRNA-seq data analysis challenges.
• Its ability to handle both clustering and batch correction enhances its applicability across various datasets.
• This framework represents a significant advancement for extracting reliable biological insights from scRNA-seq data.