Deep graph contrastive learning model for drug-drug interaction prediction

PLOS ONE RESEARCH ARTICLE Deep graph contrastive learning model for drug-drug interaction prediction Zhenyu Jiang1, Zhi Gong2,3*, Xiaopeng Dai ID1,2,3*, Hongyan Zhang1, Pingjian Ding4, Cong Shen ID5* 1 College of Information and Intelligence, Hunan Agricultural University, Changsha, China, 2 School of Computer Science and Engineering, Hunan University of Information Technology, Changsha, China, 3 Key Laboratory of Intelligent Perception and Computing, Hunan University of Information Technology, Changsha, China, 4 School of Computer Science, University of South China, Hengyang, China, 5 School of Physical and Mathematical Sciences, Nanyang Technological University, Singapore, Singapore a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Jiang Z, Gong Z, Dai X, Zhang H, Ding P, Shen C (2024) Deep graph contrastive learning model for drug-drug interaction prediction. PLoS ONE 19(6): e0304798. https://doi.org/10.1371/ journal.pone.0304798 Editor: Satyaki Roy, The University of Alabama in Huntsville, UNITED STATES Received: February 20, 2024 Accepted: May 17, 2024 Published: June 17, 2024 Copyright: © 2024 Jiang et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. * (ZG); (CS); (XD) Abstract Drug-drug interaction (DDI) is the combined effects of multiple drugs taken together, which can either enhance or reduce each other’s efficacy. Thus, drug interaction analysis plays an important role in improving treatment effectiveness and patient safety. It has become a new challenge to use computational methods to accelerate drug interaction time and reduce its cost-effectiveness. The existing methods often do not fully explore the relationship between the structural information and the functional information of drug molecules, resulting in low prediction accuracy for drug interactions, poor generalization, and other issues. In this paper, we propose a novel method, which is a deep graph contrastive learning model for drug-drug interaction prediction (DeepGCL for brevity). DeepGCL incorporates a contrastive learning component to enhance the consistency of information between different views (molecular structure and interaction network), which means that the DeepGCL model predicts drug interactions by integrating molecular structure features and interaction network topology features. Experimental results show that DeepGCL achieves better performance than other methods in all datasets. Moreover, we conducted many experiments to analyze the necessity of each component of the model and the robustness of the model, which also showed promising results. The source code of DeepGCL is freely available at https://github. com/jzysj/DeepGCL. Data Availability Statement: All data is publicly available at the following URL: https://doi.org/10. 6084/m9.figshare.25522918. Introduction Funding: The author(s) received funding for this work from the following sources: Research Foundation of Hunan Educational Committee, Grant No. 20C1579, Pingjian Ding. Hunan Province Higher Education Reform Research Project, Grant No. HNJG-2021-1242, Zhi Gong. National Natural Science Foundation of China, Grant No. 62002154, Pingjian Ding. Hunan Provincial Natural Science Drug–drug interaction (DDI) refers to the phenomenon that occurs when two or more drugs are taken together, resulting in adverse effects on an organism [1, 2]. Thus, how to accurately identify drug-drug interactions has become an important research content. Traditional methods which used in drug-drug interaction identification are mainly based on experimental assays and clinical reports [3]. However, this process would be costly and time-consuming, especially for identifying drug-drug interactions from a large drug space. Computational methods (in silico [4–6]) can be used as an effective and fast alternative to alleviate this PLOS ONE | https://doi.org/10.1371/journal.pone.0304798 June 17, 2024 1 / 18 PLOS ONE Foundation of China, Grant No. 2021JJ40467, Hongyan Zhang. Competing interests: The authors have declared that no competing interests exist. Deep graph contrastive learning model for drug-drug interaction prediction problem. Among these methods usually focus on learning single drug properties and lack effective integration of multiple sources of drug-related information, which ultimately limits the predictive capabilities of the model. Therefore, it has become an important research direction in the field of drug discovery to propose an effective and fast calculation method for drugdrug interaction prediction. In recent years, accumulated research findings have demonstrated promising results in computational-based drug-drug interactions (DDIs) prediction. These achievements are primarily attributed to the rapid advancements in drug molecular property prediction [7–10]. These methods for predicting DDIs can be broadly categorized into two groups: Structurebased methods and network-based methods. Firstly, structure-based methods mainly consider the entire drug as a graph or sequence. For example, some researchers consider atoms as nodes and bonds between atoms as edges, then use a graph neural network (GNN) to learn the representation of each drug [11–17]. Additionally, some models use SMILES (Simplified Molecular Input Line Entry System) [18] as the input for sequence models (including GRU [19], LSTM [20], and Transformer [21]), then predict the DDIs. In these methods, drugs are treated as independent individuals, and the representation is learned from the drug molecular structure and then transported to the classifier through some aggregating operations. Next, another important method for predicting DDIs is the network-based method. In this kind of method, the authors mainly consider the drug as a node, and then consider the interaction or similarity between drugs as an edge to form a large network, and then use the traditional network science method or the graph neural network method to predict the unknown interaction of drug molecules [22–24]. Although these methods have achieved good performance, they still have some limitations. Firstly, the structure-based methods assume that drugs with similar features will behave similarly in the DDIs, however, there may be a lower similarity between interacting drugs. Meanwhile, the performance of the network-based methods relies on the quality of the interaction network, and it is time-consuming and difficult to build large-scale high-quality networks. Second, the drug molecular graph and the drug interaction network contain mutually irreplaceable pharmacological properties, which are very important for predicting DDIs. The drug molecular graph contains information about the drug functional groups that determine the chemical and physical properties of the drug. The to (...truncated)