GNN-based embedding for clustering scRNA-seq data.

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) enables detailed analysis of cellular heterogeneity.
• Clustering is crucial for cell identity assignment in scRNA-seq data.
• Data sparsity and dropout events in scRNA-seq present significant clustering challenges.

Purpose of the Study:

• To introduce graph-sc, a new method for scRNA-seq data analysis using graph autoencoder networks.
• To evaluate the performance of graph-sc in clustering scRNA-seq data embeddings.
• To compare graph-sc against existing methods for cell clustering.

Main Methods:

• Development of graph-sc, a graph autoencoder network for scRNA-seq data embedding.
• Clustering of cell embeddings using various algorithms.
• Experimental validation on simulated and real scRNA-seq datasets.

Main Results:

• graph-sc demonstrates favorable performance compared to competing methods across various datasets.
• The method is stable, robust to down-sampling, and insensitive to parameter changes.
• graph-sc is computationally more efficient than other neural network-based methods.

Conclusions:

• graph-sc provides a flexible and efficient approach for scRNA-seq data analysis and cell clustering.
• The graph-based modeling allows for seamless integration of external data, such as gene networks.
• graph-sc offers a promising alternative for addressing challenges in scRNA-seq data analysis.