scMGATGRN: a multiview graph attention network-based method for inferring gene regulatory networks from single-cell...

Area of Science:

• Computational Biology
• Genomics
• Systems Biology

Background:

• Gene regulatory networks (GRNs) are crucial for understanding cellular dynamics and disease mechanisms.
• Deep learning (DL) methods have advanced GRN inference from single-cell transcriptomic data.
• Existing DL methods often overlook graph topological and high-order neighbor information.

Purpose of the Study:

• To propose a novel deep learning model, scMGATGRN, for improved GRN inference.
• To address limitations in current DL methods by incorporating multi-view graph attention.
• To enhance the utilization of graph topological and multi-order neighbor information.

Main Methods:

• Developed scMGATGRN, a multiview graph attention network model.
• Integrated Graph Attention Network (GAT) for feature extraction.
• Employed a multiview approach with view-level attention for feature aggregation.

Main Results:

• scMGATGRN demonstrated superior performance compared to 10 other methods.
• Evaluated on seven benchmark single-cell RNA sequencing (scRNA-seq) datasets.
• Outperformed shallow learning and state-of-the-art DL-based methods in GRN inference.

Conclusions:

• scMGATGRN effectively infers gene regulatory networks from scRNA-seq data.
• The model's architecture enhances the capture of complex regulatory relationships.
• scMGATGRN shows significant potential for advancing GRN research and disease mechanism studies.