Knowledge Graph-Driven AI in Biohealth

Knowledge graphs (KGs) have emerged as a powerful tool for knowledge discovery.In this perspective paper, we present a framework for KG construction, graph representation learning, and predictive modeling towards AI-driven discovery in biohealth.We illustrate this through two case studies: (1) Protein Knowledge Network (ProKN) and KSMoFinder, a KG embeddingbased model that predicts protein kinase and phosphorylation site associations with state-of-theart accuracy by learning from biological context in a biomedical knowledge network for drug discovery; (2) Social Determinants of Health (SDoH) KG, built from synthetic data of Veteran Health Administration with a veteran suicide-risk prediction model that uncovers latent, multifactorial risk patterns.These use cases spanning biomedical and population health research, demonstrate how KG-driven AI can bridge the gap between molecular and population level studies.We highlight how such open, interoperable knowledge networks offer a reusable framework for accelerating discovery and addressing complex health challenges.Finally, we provide targeted recommendations for Delaware's health innovation ecosystem to leverage this paradigm for public health strategy, clinical decision-making, and translational research. The Imperative for Connected Data in Life SciencesThe life sciences are experiencing a transformative shift, driven by the rapid expansion of data across genomics, electronic health records (EHRs), and public health domains.Artificial Intelligence (AI) has become indispensable for finding patterns in life sciences data, enabling advances in diagnostic imaging, genomic interpretation, and predictive analytics.However, conventional methods often treat data as independent features, failing to capture the rich, relational fabric of biological and health systems.Graph-based AI approaches explicitly model entities (genes, diseases, patients) and their interactions as networks, enabling the learning of higher-order dependencies.Graph representation learning has been shown to effectively harness molecular interaction networks, disease comorbidity graphs, and multimodal clinical data, providing a more holistic representation of complex biological and healthcare systems.Knowledge graphs (KGs) are semantic network structures that explicitly model entities and their relationships in a structured, queryable format, forming a basis for context-aware reasoning and inference.Early semantic integration efforts used linked data to unify diverse biological information, such as the Semantic Web for Health Care and Life Sciences (HCLS).KGs have been used to integrate knowledge from heterogeneous resources such as literature and curated databases into unified representations for disease research and discovery. 1Large-scale biomedical KGs, such as Hetionet 2 for systematic drug repurposing, may consist of an integrative network of millions of relationships among compounds, diseases, genes, pathways, and phenotypes.Recent advances explore multimodal KG constructs that support integration