An Efficient Inductive Learning Model for Inferring Gene Regulatory Networks

Identifying gene regulatory networks (GRNs) from gene expression data has been a critical focus in systems biology. This paper proposes an efficient inductive learning framework based on a directed Graph Neural Network named GRDGNN_SP. This method, a significant departure from current GNN frameworks, provides a comprehensive and explicit regulatory relationship for a GRN. By combining StructPool for hierarchical pooling, GRDGNN_SP has shown impressive results across different datasets, species, and data types for inferring GRNs, demonstrating its effectiveness. It transforms the edge prediction task in GRNs into a graph multiclassification task, where subgraphs containing target nodes and their neighbors are classified. Furthermore, by using directed graph convolutional neural networks (DGCN), our model can infer the causal relationship in gene regulation. We improve the model's performance by integrating regulatory relationships from various heuristic methods through ensemble techniques. Experimental results demonstrate that the model proposed in this paper is highly effective in inferring the causal relationships of large-scale networks from gene expression data.