Prediction of herb-drug interaction based on BioBERT

Herb-drug interaction (HDI) research plays a critical role in evaluating the safety and efficacy of clinical pharmacotherapy. Existing approaches primarily rely on molecular structure analysis or pharmacokinetic modeling, which often fail to uncover the complex mechanisms implicitly described in the literature. In this study, we propose BioBERT-HDI, a novel deep learning framework based on the pre-trained BioBERT model. Our method leverages semantic representations and an entity-focused attention mechanism to automatically predict interactions between herbal and conventional chemical drugs from unstructured text. We benchmarked BioBERT-HDI against several machine learning architectures, including K-nearest neighbors (KNN), support vector classification (SVC), random forest, BERT, and BioBERT, using a publicly available HDI dataset obtained from GitHub. Comparative analyses demonstrate that BioBERT-HDI outperforms all other models across every evaluation metric, achieving an accuracy of 95.12% and an F1 score of 0.9552. These results validate the feasibility and superiority of employing a pre-trained BioBERT framework for predicting herb-drug interactions.