Longevi graph: knowledge graphs with statistical approaches for aging

Abstract

Aging is a complex biological process influenced by various biomarkers, such as cholesterol and blood sugar levels, which serve as measurable indicators of health and disease. However, the vast amount of biomarker data presents challenges in identifying meaningful relationships. To address this, we developed Aging Biomarkers Ontology (ABO), a structured network that formally defines biomarkers from an aging perspective, organizes and connects different biomarkers, making it easier to analyze their relationships. We first applied statistical methods to detect significant differences between biomarkers across various groups. Then, we built a biomarker knowledge graph using Neo4j, a graph database that helps to represent and explore these complex relationships. To further explore aging biomarkers and provide a more complete knowledge graph, we used two approaches to uncover hidden connections: Depth-Limited Search (DLS) and Machine Learning-Based Embedding Search. The DLS approach searches the knowledge graph by traversing the nodes connected to a given node within a specified depth, whereas the embeddings-based method converts biomarker relationships into numerical representations and applies cosine similarity to predict potential associations. To compare the effectiveness of these two approaches, we analyzed how well DLS-based reasoning and embeddings-based search detected known and new relationships. By combining statistical analysis, graph-based knowledge graph completion approaches, and machine learning, this study provides a structured way to explore patterns between biomarkers. By integrating statistical analysis, knowledge graph completion approaches, and machine learning, this study provides a structured framework for aging biomarker analysis, improving data exploration, and aid biomedical research in longevity and aging.