Biomedical texts provide important data for investigating drug-drug interactions (DDIs) in the field of pharmacovigilance. Although researchers have attempted to investigate DDIs from biomedical texts and predict unknown DDIs, the lack of accurate manual annotations significantly hinders the performance of machine learning algorithms. In this study, a new DDI prediction framework, Subgraph Enhance model, was developed for DDI (SubGE-DDI) to improve the performance of machine learning algorithms. This model uses drug pairs knowledge subgraph information to achieve large-scale plain text prediction without many annotations. This model treats DDI prediction as a multi-class classification problem and predicts the specific DDI type for each drug pair (e.g. Mechanism, Effect, Advise, Interact and Negative). The drug pairs knowledge subgraph was derived from a huge drug knowledge graph containing various public datasets, such as DrugBank, TwoSIDES, OffSIDES, DrugCentral, EntrezeGene, SMPDB (The Small Molecule Pathway Database), CTD (The Comparative Toxicogenomics Database) and SIDER. The SubGE-DDI was evaluated from the public dataset (SemEval-2013 Task 9 dataset) and then compared with other state-of-the-art baselines. SubGE-DDI achieves 83.91% micro F1 score and 84.75% macro F1 score in the test dataset, outperforming the other state-of-the-art baselines. These findings show that the proposed drug pairs knowledge subgraph-assisted model can effectively improve the prediction performance of DDIs from biomedical texts.

Background: Drug–drug interactions (DDIs) account for about 30% of adverse drug reactions and 5–10% of hospital deaths. Combination therapy increases DDI risks, yet extracting DDIs from biomedical text remains challenging: existing methods rely on surface co-occurrence and fail when multiple drugs and interactions coexist in a sentence. Prior multimodal approaches simply concatenate text, molecular, or knowledge features without deep alignment, leading to misclassification of structurally similar but non-interacting drug pairs. Methods: We propose MultiMod-DDI, a framework that constructs a ternary evidence chain of “molecular structure–biological entities–DDI text”. Unlike existing work, MultiMod-DDI introduces (1) PS-AEGNN, a molecular graph network with ProbSparse self-attention to capture long-range chemical dependencies; (2) an adaptive position interaction vector that dynamically weights distant semantic links between drug entities; and (3) a multi-stage adaptive fusion module that sequentially applies subgraph-molecule attention and text-guided gating. These components are co-designed to enforce structured semantic alignment among heterogeneous modalities, effectively addressing the specific challenge of matching drug pairs to their correct interaction types in complex, multi-drug sentences. Results: On SemEval-2013 Task 9, MultiMod-DDI achieves 85.57% F1_macro and 85.20% F1_micro, outperforming state-of-the-art models. Conclusions: Through multimodal deep semantic alignment, MultiMod-DDI effectively resolves the mismatch between drug pairs and their interaction types in complex biomedical texts. The integration of multimodal features greatly improves DDI extraction accuracy, offering a reliable method for intelligent DDI mining from biomedical literature.