Inferring gene regulatory networks by hypergraph generative model.

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) generates high-dimensional data crucial for understanding cellular heterogeneity.
• Inferring gene regulatory networks (GRNs) from scRNA-seq data is challenging due to data complexity and noise.
• Existing models often struggle to capture intricate biological relationships and cellular variations effectively.

Purpose of the Study:

• To develop a novel Bayesian deep generative model, HyperG-VAE, for enhanced scRNA-seq data analysis.
• To improve the inference of gene regulatory networks (GRNs) and facilitate single-cell clustering.
• To leverage hypergraph representations for a more comprehensive understanding of cellular heterogeneity and gene interactions.

Main Methods:

• Developed HyperG-VAE, a hypergraph variational autoencoder model.
• Incorporated a structural equation model in the cell encoder to address cellular heterogeneity and construct GRNs.
• Utilized hypergraph self-attention in the gene encoder to identify gene modules.
• Employed synergistic optimization of encoders and a decoder for improved GRN inference and clustering.

Main Results:

• HyperG-VAE demonstrated improved GRN inference, single-cell clustering, and data visualization capabilities, validated by benchmarks.
• The model effectively uncovered gene regulation patterns in B cell development data.
• Gene set enrichment analysis confirmed the gene encoder's effectiveness in refining GRN inference.

Conclusions:

• HyperG-VAE provides an efficient and robust solution for scRNA-seq analysis and GRN construction.
• The model successfully captures cellular heterogeneity and identifies gene modules.
• HyperG-VAE shows potential for future applications in temporal and multimodal single-cell omics data analysis.