Revealing hidden regulatory dependencies: multi-perspective graph learning for single-cell gene regulatory network...

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Gene regulatory networks (GRNs) are crucial for understanding cellular processes and diseases.
• Accurate GRN inference from single-cell transcriptomic data is challenging due to data complexity.
• Integrating multi-level gene features is key to improving GRN inference.

Purpose of the Study:

• To develop an advanced framework, ATFGRN, for accurate gene regulatory relationship prediction.
• To effectively integrate diverse gene features for enhanced GRN inference.
• To improve the accuracy of gene regulatory network construction from single-cell data.

Main Methods:

• ATFGRN utilizes an adaptive topology-feature fusion graph neural framework.
• It incorporates subgraph structure encoding, expression-guided graph convolutional networks with self-attention, and a KNN graph with graph attention for similarity.
• Features from these modules are fused using an attention-based weighting mechanism.

Main Results:

• ATFGRN demonstrated improved performance in predicting gene regulatory relationships.
• Evaluations on single-cell transcriptomic datasets showed a 5.09% increase in AUROC performance compared to existing methods.
• The multi-perspective fusion strategy proved effective and applicable.

Conclusions:

• ATFGRN offers a powerful approach for accurate GRN inference.
• The integration of structural, expression, and similarity perspectives enhances prediction capabilities.
• This framework advances the analysis of gene regulation using single-cell transcriptomic data.