Anomaly Detection with Graph Embedding in Bioinformatics: A Survey

With the rapid development of network and data sciences, graph anomaly detection has emerged as a key research direction across multiple domains, and its application has been widely recognized in bioinformatics. Biomolecular networks, including gene regulatory networks, protein-protein interaction networks, and drug-molecule interaction networks, form the foundation of complex biological processes within organisms. Abnormal patterns within these networks are often closely associated with the onset and progression of diseases, making graph anomaly detection critically important for revealing disease mechanisms, facilitating disease diagnosis, and advancing drug development. In this paper, we review the current status of graph embedding-based anomaly detection algorithms in bioinformatics, with a particular focus on key literature and the development of dynamic graph anomaly detection algorithms. We explore how graph embedding techniques learn high-dimensional features of biomolecular networks and represent them in a low-dimensional space. Furthermore, we investigate the dynamic nature of biomolecular networks and how graph embedding algorithms can capture these changes to improve detection accuracy and efficiency. The challenges of current methods are summarized, and future directions are proposed to support key applications such as disease diagnosis, gene function annotation, and drug discovery. Through this review, we aim to provide theoretical foundations and methodological support for network anomaly detection in high-throughput omics data.