SIGRN: Inferring Gene Regulatory Network with Soft Introspective Variational Autoencoders.

Area of Science:

• Computational Biology
• Genomics
• Systems Biology

Background:

• Gene regulatory networks (GRNs) are crucial for understanding biological processes.
• Inferring GRNs from single-cell RNA sequencing (scRNA-seq) data is computationally challenging.
• Existing methods like Variational Autoencoders (VAEs) have limitations in data quality.

Purpose of the Study:

• To develop an improved computational method for inferring GRNs from scRNA-seq data.
• To enhance the accuracy and quality of GRN inference and scRNA-seq data generation.
• To address the data quality shortcomings of VAE-based approaches.

Main Methods:

• Proposed SIGRN (soft introspective gene regulatory network) model.
• Introduced an adversarial mechanism within a VAE framework.
• Employed a "soft" introspective adversarial mode to optimize model parameters efficiently.

Main Results:

• SIGRN demonstrated superior inference accuracy compared to nine leading methods on benchmark datasets.
• Achieved better performance in cell representation and scRNA-seq data generation.
• Experimental validation confirmed the method's effectiveness.

Conclusions:

• SIGRN offers a promising approach for accurate GRN inference.
• The method shows potential for improving scRNA-seq data generation.
• SIGRN advances computational tools for systems biology research.