GCLink: a graph contrastive link prediction framework for gene regulatory network inference.

Area of Science:

• Computational Biology
• Genomics
• Systems Biology

Background:

• Gene regulatory networks (GRNs) are crucial for understanding cellular processes.
• Single-cell RNA sequencing (scRNA-seq) allows GRN inference at single-cell resolution.
• Existing methods often predict pairwise interactions, limiting comprehensive network analysis and generalization.

Purpose of the Study:

• To propose a novel model, GCLink, for inferring gene regulatory interactions from scRNA-seq data.
• To enhance the prediction of potential gene regulatory interactions by leveraging graph contrastive learning.
• To improve the generalization performance of GRN inference, particularly in data-limited scenarios.

Main Methods:

• Developed a graph contrastive link prediction (GCLink) model.
• Utilized a graph contrastive learning strategy to aggregate gene feature and neighborhood information.
• Trained and evaluated the model on real scRNA-seq datasets, including pretraining and fine-tuning approaches.

Main Results:

• GCLink effectively infers potential gene regulatory interactions from scRNA-seq data.
• The model demonstrates superior performance compared to state-of-the-art methods on real datasets.
• GCLink shows strong performance in GRN inference even with limited known interactions, highlighting its generalization capability.

Conclusions:

• GCLink offers an effective approach for inferring gene regulatory networks from scRNA-seq data.
• The graph contrastive learning strategy enhances the accuracy and robustness of network inference.
• The model's ability to perform well with limited prior knowledge makes it valuable for diverse biological applications.