Aug 2026· IEEE journal of biomedical and health informatics· Vol PP, pp. 1-10· 0 citations
Medicine
TL;DR
Experimental results show that CMAF-DDI improves multi-class DDI prediction compared with representative graph-based and multi-source fusion baselines and ablation, hyperparameter sensitivity, controlled protein perturbation, representation, and case-level analyses to examine the contribution and behavior of protein-enhanced cross-modal fusion.
Abstract
Accurate prediction of drug-drug interactions (DDIs) is crucial for medication safety and personalized treatment. Most existing methods primarily exploit molecular graphs or biomedical knowledge graphs, while target protein sequence information is often underused. This paper proposes CMAF-DDI, a multi-class DDI prediction framework that integrates protein sequence features, molecular graph features, and knowledge graph features. CMAF-DDI contains a bi-level cross-modal fusion module: an Attention Fusion (AF) level that models global dependencies among modalities using multi-head attention, and a Triple-feature Product Fusion (TPF) level that captures high-order cross-modal co-activation after projecting all modalities into a shared latent space. Experimental results on DrugBank and DRKG show that CMAF-DDI improves multi-class DDI prediction compared with representative graph-based and multi-source fusion baselines. We further provide ablation, hyperparameter sensitivity, controlled protein perturbation, representation, and case-level analyses to examine the contribution and behavior of protein-enhanced cross-modal fusion.
Experiments demonstrate that HSAF-DDI achieves superior overall performance compared to state-of-the-art methods, indicating the critical role of fine-grained biological features in improving the DDI prediction performance.
Xiaoli Lin, Si-Yuan Zhang, Bo Li et al.· Journal of Chemical Informat...· 0 citations
A novel multiview feature fusion-based graph representation model (MFF-GRM) for predicting DDI that integrates drug molecular graphs, SMILES sequences, DDI information networks, and drug biological features to learn drug features more comprehensively.
Mengyuan Jin, Dan Liu, E. Benfenati et al.· Applied intelligence (Boston...· 0 citations
Biomedical drug discovery spans heterogeneous modalities-molecular structures, knowledge graphs(KG), structure-aware features (3D molecular conformation and protein physicochemical geometry), and textual annotations-yet existing methods suffer from modality isolation, incomplete relational supervision, and uneven per-sample reliability, limiting generalization to unseen compounds and targets. We propose UniDrug-LLM, a unified reliability-aware multi-modal fusion framework covering four core tasks: drug-target, drug-drug, and protein-protein interaction prediction(DTI, DDI, and PPI), and drug property estimation(DP). By performing integrated representation learning across multiple modalities, our model breaks the isolation of single-modal paradigms. Specifically, we propose Cross-Modal Subspace Alignment (CMSA), which treats incomplete knowledge graphs as recoverable signals, synthesizing missing relational embeddings via cross-modal retrieval and subspace generation. We further design the Cross-Modal Orthogonal Decomposition Network (CMON), which decomposes features into shared and private components, estimates per-sample modality reliability, and emits digest tokens-compact tokens that summarize each modality's reliability for the backbone-to enable adaptive fusion. Both modules are integrated through a LoRA-tuned LLM backbone for end-to-end prediction. Extensive experiments show that UniDrug-LLM matches or exceeds state-of-the-art baselines, with improvements of up to 8.8%, and it generalizes well to cold-start settings, where drugs or targets are unseen during training, in the DTI task. This work establishes a principled paradigm shift toward reliability-aware multi-modal learning for drug discovery.
Xiaodong Zhu, Longjun Song, Shiying Zeng et al.· IEEE journal of biomedical a...· 0 citations
Drug-drug interaction (DDI) prediction is an important task in computational pharmacology because unidentified interactions may reduce therapeutic efficacy or induce severe adverse effects. Although recent deep learning methods have achieved promising performance, most existing approaches primarily rely on either two-dimensional (2D) molecular topology or coarse multimodal fusion strategies, while insufficiently modeling fine-grained interaction dependencies between drug pairs. To address these limitations, we propose IAMV-DDI, an interaction-aware multi-view molecular representation learning framework for DDI prediction and DDI event classification. The proposed framework jointly integrates 2D molecular topology, 3D spatial geometry, and token-level interdrug interaction modeling. Specifically, a SimSGT-based masked graph encoder is employed to learn informative 2D molecular representations, while an E(n) Equivariant Graph Neural Network (EGNN) encoder with contrastive conformer pretraining captures geometry-aware 3D structural features. The learned 2D and 3D token representations are integrated through a gated cross-modal fusion module, followed by a bidirectional cross-attention mechanism to explicitly model interaction-aware dependencies between drug pairs. Experiments conducted on the benchmark DrugBank and ZhangDDI data sets demonstrate that IAMV-DDI achieves strong performance compared with representative network-based, chemical-structure-based, and hybrid baseline methods. In binary DDI prediction, IAMV-DDI achieves highly competitive performance on DrugBank and ZhangDDI, with closely matched results to the strongest baseline on ZhangDDI. In DrugBank multiclass DDI event classification, IAMV-DDI achieves an Accuracy of 0.9650, Macro-Precision of 0.9439, Macro-Recall of 0.9347, and Macro-F1 of 0.9361, substantially outperforming the strongest baseline. Ablation studies further confirm the effectiveness of the multiview molecular fusion strategy and the interaction-aware cross-attention mechanism. These results demonstrate that jointly modeling molecular topology, spatial geometry, and fine-grained interdrug dependencies can produce highly discriminative representations for accurate DDI prediction.
Huyen K. Nguyen, Quang H. Nguyen, D. Le· Journal of Chemical Informat...· 0 citations
DDI-MMAF is proposed, a lightweight cross-modal framework that avoids using explicit high-dimensional omics profiles as direct model inputs and enables accurate synergy prediction from raw minimalist inputs, offering a practical and efficient computational solution for cost-effective drug combination discovery.
Hao Li, Qianhui Jiang, Jiahui Guan et al.· European journal of medicina...· 0 citations
Results indicate that integrating heterogeneous structural cues through coarse- and fine-grained feature interaction provides an effective and scalable solution for DDI prediction.
