Inductive inference of gene regulatory network using supervised and semi-supervised graph neural networks

Area of Science:

• Bioinformatics
• Computational Biology
• Systems Biology

Background:

• Reconstructing gene regulatory networks (GRNs) from gene expression data is a fundamental challenge in bioinformatics.
• Inferring regulatory relationships can be framed as a link prediction problem in graph theory.

Purpose of the Study:

• To propose an end-to-end Gene Regulatory Graph Neural Network (GRGNN) for reconstructing GRNs from scratch using gene expression data.
• To enhance GRN inference through supervised and semi-supervised learning frameworks.

Main Methods:

• Formulating GRN inference as a graph classification problem to predict links between transcription factors (TFs) and target genes.
• Utilizing a Graph Neural Network (GNN) with node features from gene expression and graph embeddings.
• Incorporating noisy initial graph structures (e.g., from Pearson correlation) to guide inference via ensemble techniques.
• Implementing a semi-supervised scheme to improve classifier performance.

Main Results:

• The GRGNN approach demonstrated superior performance in reconstructing GRNs.
• Achieved state-of-the-art results on the DREAM5 GRN inference benchmarks.
• The method effectively integrates topological information and gene expression data.

Conclusions:

• GRGNN offers a powerful and accurate method for de novo GRN reconstruction.
• The proposed graph classification framework enhances the generalization capability of GNNs for GRN inference.
• GRGNN provides a valuable tool for understanding gene regulation and is publicly available.