scBiG for representation learning of single-cell gene expression data based on bipartite graph embedding.

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Single-cell RNA sequencing (scRNA-seq) data analysis is challenging due to high dimensionality, sparsity, and noise.
• Dimensionality reduction is crucial for simplifying scRNA-seq data complexity and improving signal-to-noise ratio.

Purpose of the Study:

• Introduce scBiG, a novel graph node embedding method for scRNA-seq data representation learning.
• Evaluate scBiG's performance against existing dimensionality reduction techniques in various analytical tasks.

Main Methods:

• scBiG constructs a bipartite graph connecting cells and expressed genes.
• A multilayer graph convolutional network is employed to learn cell and gene embeddings.
• The method was tested on downstream tasks including cell clustering, trajectory inference, and gene expression analysis.

Main Results:

• scBiG demonstrates superior performance compared to commonly used dimensionality reduction techniques.
• The method shows effectiveness in unsupervised cell clustering and cell trajectory inference.
• scBiG offers computational efficiency and scalability for large scRNA-seq datasets.

Conclusions:

• scBiG provides a powerful graph neural network framework for scRNA-seq data representation learning.
• The method facilitates diverse downstream analyses, including gene expression and co-expression analysis.
• scBiG enhances the interpretability and utility of scRNA-seq data.