A comprehensive survey on graph neural networks for gene regulatory network inference.

The gene regulatory network (GRN) represents a complex web of genetic interactions that governs cellular functions and responses to environmental stimuli. Understanding these intricate relationships is crucial for advancing developmental biology, disease modeling, and therapeutic discovery. With the growing interest in graph-based approaches, graph neural networks (GNNs) have emerged as a powerful tool for GRN inference, offering the ability to capture high-dimensional dependencies and topological structures within gene networks. This survey presents the first comprehensive review of GNN-based methods for GRN inference, analyzing 16 state-of-the-art approaches. We categorize these methods based on their underlying architectures, inference strategies, and computational frameworks. Additionally, we provide a critical evaluation of their strengths, limitations, and real-world applicability. Unlike prior surveys that focus on either scRNA-seq or deep learning broadly, this work systematically unifies graph architectures, learning paradigms, and data regimes under a common benchmarking framework. By identifying key challenges-such as scalability, interpretability, and dataset limitations-this survey aims to guide both life scientists in selecting appropriate computational models and researchers in developing next-generation GRN inference techniques using graph-based learning.