AutoGRN: An Automated Graph Neural Network Framework for Gene Regulatory Network Inference.

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Gene regulatory network (GRN) inference is crucial for understanding biological processes and diseases.
• Single-cell RNA sequencing (scRNA-seq) data is widely used for GRN inference but presents challenges like sparsity, noise, and heterogeneity.
• Existing Graph Neural Network (GNN) approaches struggle to generalize across diverse scRNA-seq datasets due to fixed architectures.

Purpose of the Study:

• To develop an automated framework, AutoGRN, for optimizing GNN architectures for GRN inference.
• To address the limitations of fixed GNN architectures in handling the complexities of scRNA-seq data.
• To enhance the accuracy and generalizability of GRN inference methods.

Main Methods:

• AutoGRN employs a genetic search algorithm, constrained by information entropy, to explore and identify optimal GNN architectures.
• The framework incorporates various components influencing GRN inference performance into its search space.
• The automated approach adapts GNN architectures to the specific characteristics of individual scRNA-seq datasets.

Main Results:

• AutoGRN demonstrates superior prediction accuracy compared to existing GRN inference methods.
• The proposed framework shows robust performance across multiple public scRNA-seq datasets.
• The automated architecture search effectively handles data sparsity, noise, and heterogeneity.

Conclusions:

• AutoGRN provides an effective and adaptable solution for gene regulatory network inference.
• The automated GNN architecture optimization significantly improves GRN inference accuracy and generalizability.
• This framework advances the analysis of gene interactions from complex single-cell expression data.