AHormoneDB: A cross-species hormone knowledge base with AI-powered target prediction and natural-language querying

Hormones regulate growth, development, metabolism, reproduction, and stress responses, yet data required to study their regulatory networks remain scattered across separate databases and analytical tools. AHormoneDB unifies these cross-species resources by curating standardized hormone–receptor–gene–disease relationships and providing a user-friendly analytical environment.

The platform integrates 7,833 standardized hormone–gene–disease association records covering 367 standardized hormone labels. Its GeneReg v2.6 module combines evidence from ChIP-seq peaks, motif scanning, and literature co-occurrence. Evaluated on an internal test set, GeneReg achieved an F1 score of 0.894, an AUC of 0.923, and an AUPRC of 0.924. These metrics represent internal benchmark results, and all predicted relationships require experimental validation.

AHormoneDB also features a retrieval-augmented generation assistant for exploratory, natural-language research questions. Its hybrid keyword-and-semantic retrieval pipeline attained an F1 score of 0.849 on a 200-question test set. RAGAS, an evaluation framework for retrieval-augmented generation, measured a faithfulness score of 0.900 and answer relevance of 0.883. In an acromegaly case study, the platform recapitulated the GH–GHR–IGF1 axis and JAK–STAT and PI3K–AKT signaling cascades, and nominated PTPN1 as a candidate node for further investigation. The platform provides a no-code interface for regulatory network visualization and drug-target mining. AHormoneDB is accessible at https://panhorm.cn, and its source code is available at https://github.com/ouev5/panhorm.