Graph neural networks accelerate drug discovery for complex bone diseases

Announcing a new publication for Acta Materia Medica journal. Traditional Chinese medicine has shown therapeutic potential in treating osteoarthritis (OA) by regulating inflammation and maintaining cartilage homeostasis.

However, the complex compositions of herbal medicines and the lack of efficient screening strategies have hindered the identification of active compounds and their molecular mechanisms. To address these challenges, this study used graph neural networks (GNNs) for drug discovery and demonstrated their potential in elucidating therapeutic mechanisms. Using an in-house GNN model, we identified Paederia scandens as a promising candidate for OA treatment. Experimental validation confirmed that Paederia scandens improved cartilage metabolic homeostasis and mitigated subchondral bone sclerosis.

Further analysis implicated asperuloside, a major constituent of Paederia scandens, as a key bioactive compound contributing to these therapeutic effects. Transcriptomic profiling and protein-protein interaction network analysis identified Integrin Subunit Beta 1 as a potential central regulatory hub. Asperuloside treatment was found to reshape cartilage gene expression; downregulate cytokine and chemokine signaling pathways; and alleviate inflammation, while enhancing cartilage matrix synthesis and decreasing matrix degradation. Further in vivo and in vitro experiments consistently supported these findings. Collectively, our findings indicated that asperuloside is a promising OA therapeutic candidate. Moreover, the GNN-driven framework established in this study provides a novel strategy for modernizing traditional Chinese medicine and accelerating the discovery of bioactive compounds.

This work highlights the critical role of GNNs in integrating computational prediction with biological validation to facilitate mechanistic exploration and advance precision drug development for complex diseases such as OA.