GMFGRN: a matrix factorization and graph neural network approach for gene regulatory network inference

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) enables gene expression profiling at the individual cell level.
• Accurate inference of gene regulatory networks (GRNs) from scRNA-seq data is crucial for understanding cellular mechanisms.
• Existing GRN inference methods often struggle with transitive interactions and high computational costs.

Purpose of the Study:

• To develop a novel, accurate, and efficient method for GRN inference from scRNA-seq data.
• To address limitations of existing methods, including transitive interactions and computational resource demands.

Main Methods:

• Introduced GMFGRN, a graph neural network (GNN)-based method for GRN inference.
• Employed GNN for matrix factorization to learn gene embeddings.
• Utilized learned embeddings to predict transcription factor-gene interactions.

Main Results:

• GMFGRN demonstrated superior performance on eight static scRNA-seq datasets, showing mean improvements of 1.9% (AUROC) and 2.5% (AUPRC) over the runner-up.
• On four time-series datasets, GMFGRN achieved maximum enhancements of 2.4% (AUROC) and 1.3% (AUPRC) compared to the runner-up.
• GMFGRN required significantly less training time and memory (<10% of the second-best method).

Conclusions:

• GMFGRN offers a substantial advancement in GRN inference from scRNA-seq data.
• The method provides a computationally efficient and accurate alternative to existing approaches.
• GMFGRN has significant potential for advancing systems biology research.