Title: Enhancing drug-drug interaction event prediction from knowledge graphs by multimodal deep neural networks

Authors: Xiaomin Shen; Jianliang Gao; Tengfei Lyu; Jiamin Chen; Jiarun Zhang; Jing He; Zhao Li; Wei Yu

Addresses: Collaborative Innovation Center for Diagnosis and Treatment of Infectious Diseases, State Key Laboratory for Diagnosis and Treatment of Infectious Diseases, The First Affiliated Hospital, College of Medicine, Zhejiang University, Hangzhou, China ' School of Computer Science and Engineering, Central South University, Changsha, China ' School of Computer Science and Engineering, Central South University, Changsha, China ' School of Computer Science and Engineering, Central South University, Changsha, China ' University of California, San Diego, USA ' Nuffield Department of Clinical Neurosciences, University of Oxford, Oxford, UK ' Collaborative Innovation Center for Diagnosis and Treatment of Infectious Diseases, State Key Laboratory for Diagnosis and Treatment of Infectious Diseases, The First Affiliated Hospital, College of Medicine, Zhejiang University, Hangzhou, China ' Collaborative Innovation Center for Diagnosis and Treatment of Infectious Diseases, State Key Laboratory for Diagnosis and Treatment of Infectious Diseases, The First Affiliated Hospital, College of Medicine, Zhejiang University, Hangzhou, China

Abstract: This study tackles the challenging issue of predicting drug-drug interactions (DDI) in pharmacology. Despite advancements in deep learning for DDI prediction, many techniques fail to fully leverage multimodal data correlations, limiting accuracy. To address this, we propose the knowledge graphs by multimodal deep neural network (KGMDNN) framework, enhancing DDI prediction by integrating features from drug knowledge graphs (DKG) and heterogeneous features (HF). KGMDNN uses a dual-path structure to obtain multimodal drug representations, effectively capturing drug relationships and connections within DKG to improve prediction accuracy. Our method excels in learning joint representations of structural information and multimodal data, as demonstrated through numerous real-world dataset experiments. Additionally, testing various drug knowledge graphs confirmed the model's robustness. KGMDNN outperforms both classic and state-of-the-art models in prediction metrics and interpretability.

Keywords: DDI event prediction; drug-drug interaction; graph neural network; GNN; knowledge graph; heterogeneous information; multi-modal data.

DOI: 10.1504/IJDMB.2025.147056

International Journal of Data Mining and Bioinformatics, 2025 Vol.29 No.3, pp.338 - 361

Received: 24 May 2023
Accepted: 11 Jan 2024
Published online: 10 Jul 2025 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article