Global knowledge graph of osteoporosis biomarkers based on large language model embeddings and complex network algorithms.

BACKGROUND

This study applies advanced artificial intelligence technologies to reconstruct the global research landscape of osteoporosis (OP) biomarkers and to predict emerging nodes with potential clinical value.

METHODS

Literature from the Web of Science Core Collection and PubMed over the past 20 years was integrated. The OpenAI text-embedding-3-large model mapped keywords into a high-dimensional semantic space, enabling deep clustering through cosine similarity and reducing polysemy. The Walktrap community detection algorithm and Callon strategic diagram were used to identify high-centrality, low-density themes in the "fourth quadrant." Gemini 2.5 Pro was applied to generate entity-relation triples, which were validated through a human-in-the-loop process and integrated into a refined knowledge graph. Topological indicators such as betweenness centrality were used to detect emerging potential nodes.

RESULTS

A total of 1595 core publications were included, showing a significant linear rise in annual output. Nine major research themes were identified. The strategic diagram highlighted three high-potential domains: inflammation and immune regulation, detection technologies and biosensing, and gut microbiota and metabolism. Knowledge-graph analysis revealed oxidative stress as a key bridging node integrating immune, metabolic, and other systems. Several emerging nodes with strong connective potential were identified outside traditional hotspots, including molecular biomarkers such as lncRNA MEG3, miR-21-5p, spermidine, and Treg-Exo, as well as imaging innovations such as opportunistic CT screening and vertebral bone marrow PDFF. These findings indicate a paradigm shift in OP research from single-hormone mechanisms toward a multidimensional network involving inflammation, immunity, gut microbiota, and imaging.

CONCLUSION

Integrating AI-based embedding models with complex network algorithms, this study overcomes limitations of traditional bibliometrics and systematically uncovers the latent knowledge structure of OP biomarkers. The identified emerging nodes offer robust evidence and forward-looking guidance for early precision screening-such as liquid biopsy and radiation-free imaging-and for developing novel therapeutic strategies, including bone-targeted exosomes and gut-bone axis interventions.