GRLGRN: graph representation-based learning to infer gene regulatory networks from single-cell RNA-seq data

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Gene regulatory networks (GRNs) model gene interactions crucial for cellular functions.
• Single-cell RNA sequencing (scRNA-seq) offers insights but presents challenges like noise and dropout for GRN reconstruction.
• Existing machine learning and deep learning methods struggle with scRNA-seq data complexity.

Purpose of the Study:

• To develop a novel deep learning model, GRLGRN, for accurate GRN inference.
• To leverage prior GRN knowledge and scRNA-seq data for improved gene regulatory relationship prediction.
• To address challenges posed by cellular heterogeneity and data imperfections in scRNA-seq.

Main Methods:

• GRLGRN employs a graph transformer network to capture implicit regulatory links from a prior GRN.
• Gene features are encoded using an adjacency matrix of implicit links and gene expression profiles.
• Attention mechanisms enhance feature extraction, refining gene embeddings for relationship inference.

Main Results:

• GRLGRN outperformed prevalent models in predicting gene interactions across seven cell-line datasets.
• Achieved superior performance in AUROC (Area Under the Receiver Operating Characteristic Curve) and AUPRC (Area Under the Precision-Recall Curve) on most datasets.
• Demonstrated significant average improvements: 7.3% in AUROC and 30.7% in AUPRC.

Conclusions:

• GRLGRN shows strong performance in predicting gene interactions from scRNA-seq data.
• The model provides interpretability, enabling identification of key regulatory genes (hub genes) and uncovering hidden regulatory links.
• Results highlight GRLGRN's potential for advancing biological research through accurate GRN reconstruction.