Predicting gene regulatory links from single-cell RNA-seq data using graph neural networks.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) enables gene expression analysis at the individual cell level.
• Inferring gene regulatory networks (GRNs) from scRNA-seq data is crucial for understanding cellular functions but is hindered by data sparsity and noise.
• Supervised methods for GRN inference are increasingly feasible due to abundant transcription factor binding data.

Purpose of the Study:

• To develop a novel framework, GNNLink, for accurate gene regulatory network inference from scRNA-seq data.
• To address the challenges of sparsity, noise, and dropout events in scRNA-seq data for GRN reconstruction.
• To leverage graph link prediction and graph convolutional networks for enhanced GRN inference.

Main Methods:

• scRNA-seq data preprocessing.
• Utilizing a graph convolutional network-based interaction graph encoder to refine gene features by capturing network interdependencies.
• Employing matrix completion on node features for GRN inference.
• Framing GRN inference as a graph link prediction task.

Main Results:

• GNNLink demonstrates comparable or superior performance against six existing GRN reconstruction methods across seven diverse scRNA-seq datasets.
• The framework shows robustness and accuracy across various ground truth network types (functional interactions, LoF/GoF, ChIP-seq).
• Consistent performance was observed across scRNA-seq datasets of varying scales.

Conclusions:

• GNNLink offers a robust and accurate approach for gene regulatory network inference from challenging scRNA-seq data.
• The developed method can be applied to downstream tasks like measuring gene pair similarity and inferring causality.
• The framework's effectiveness is validated across multiple datasets, highlighting its potential for biological discovery.