Code for the paper named "Inferring gene regulatory networks by hypergraph variational autoencoder".

In constructing Gene Regulatory Networks (GRN), it is vital to account for both differential gene module regulation and the diversity stemming from cellular heterogeneity. Here we present HyperG-VAE, a Bayesian deep generative model that utilizes a hypergraph for analyzing single-cell RNA sequencing (scRNA- seq) data. HyperG-VAE employs a cell encoder with a Structural Equation Model for GRN construction and cellular heterogeneity analysis, alongside a gene encoder using hypergraph self-attention for gene module identification, cohesively optimized by a decoder. HyperG-VAE excels in GRN inference and scRNA-seq visualization, confirmed by benchmarks. It effectively reveals gene regulation in bone marrow B cell development and shows robustness in varied downstream analyses. The model’s cell encoder identifies distinct cell clusters and maps B cell progression. As another key contribution, gene set enrichment analysis (GSEA) validates the gene encoder’s module associations with specific cellular contexts, demonstrating its comprehensive analytical capabilities.

• Pytorch 1.12.1
• Python 3
• CUDA 11.6
• Different environment setting may make very subtle difference

Command to run HyperG-VAE

• Gene Regulation Inference (including cell type specific GRN and cell type non-specific GRN).

    python mainspec.py --task celltype_GRN --data_file <scGNA-seq path> --save_name <output path> --setting test
    python main.py --task non_celltype_GRN --data_file <scGNA-seq path> --save_name <output path> --setting test

  use --setting test to infer GRN instead of benchmarking.
• Embedding

    python mainumap.py --task non_celltype_GRN --data_file <scGNA-seq path> --save_name <output path>

Hyperparameters tuned in the paper can be found in the Supplementary.

• The basic setting and datasets used in the scRNA-seq GRN inference benchmark part could be found in "BEELINE".
• EPR and AUPRC calculation tutorial could be found in "DeepSEM".
• SCENIC tutorial can be found in "SCENIC".
• GSEA analysis tool can be found "here".
• Chip-Atlas dataset can be found "here".
• scRNA-seq data with clustering label annotation preprocess tutorial can be found "here".
• GRN networks are drawn by Cytoscape.