A systems-level dissection of aging and longevity mechanisms powered by an evidence-routed AI knowledge graph

Abstract

Abstract Aging is a complex biological process influenced by numerous molecular and cellular factors. A systems-level understanding of these mechanisms is essential for identifying key regulatory processes and prioritizing effective interventions. Here, we present a comprehensive dissection of aging and longevity mechanisms powered by an evidence-routed AI-based knowledge graph, HALDxAI. We integrated 445,435 aging- and longevity-related PubMed records with 19 curated biomedical databases to construct a unified Aging Knowledge Graph (Aging-KG). The pipeline combined LLM extraction from two selected abstract subsets with machine learning and deep learning extraction at corpus scale. HALDxAI uses a multi-agent retrieval system that links graph relations and generated answers to sentence-level evidence. In an expert review of 300 database relations, all cited PMIDs resolved, 97.3% of the displayed evidence was attributable to the cited record, and 65.3% of the relation claims were fully supported by that evidence. Network analyses of the Aging-KG identified TP53, mTOR, and SIRT1 as highly ranked bridge candidates between aging and longevity seed sets. These rankings describe graph topology and do not establish regulatory function or causality. In addition, an inflammaging-focused case study demonstrates the capacity of HALDxAI to decompose modular structures and position anti-aging interventions. Overall, HALDxAI reduces knowledge fragmentation and provides a scalable platform for evidence retrieval, network analysis, and hypothesis generation in aging and longevity research.