Terpenoids exhibit diverse biological activities and thus have a wide range of pharmacological applications. In modern drug discovery, data-driven deep models play a crucial role in facilitating efficient feature representation and knowledge inference. To explore the uncharted bioactivity space of terpenoids, the construction of a multi-dimensional relational terpenoid database is essential for mapping terpenoid-bioactivity profiles. In this study, we first constructed a large-scale biological knowledge graph by integrating various data types, including terpenoid compounds, protein targets, cellular targets, genes, diseases, and their interrelationships. Subsequently, we developed a network-based disease prediction model, as well as optimized multiple compound-protein interaction prediction tools to extend the framework for activity research. These resources have been deployed on a user-friendly web platform (TeroACT) accessible at: http://terokit.qmclab.com/teroact/. Using in silico models within the TeroACT platform, we screened multiple terpenoid molecules for anti-melanoma activity. In vitro and in vivo animal models further validated the anti-migration and anti-proliferative effects of mollugin and columbianadin in melanoma. Additionally, integrated computational screening and experimental approaches identified numerous terpenoids with anti-inflammatory properties. In this sense, TeroACT fills the gap in terpenoid bioactivity study by providing a comprehensive data resource and AI-driven drug discovery tools.

Terpenoids  /  Knowledge graph  /  Deep learning  /  Disease prediction  /  Drug reposition  /  Deep neural network  /  Anti-melanoma  /  Anti-inflammatory