The study of molecule-target interaction is quite important for drug discovery in terms of target identification, hit identification, pathway study, drug-drug interaction, etc. Most existing methodologies utilize either biomedical network information or molecule structural features to predict potential interaction link. However, the biomedical network information based methods usually suffer from cold start problem, while structure based methods often give limited performance due to the structure/interaction assumption and data quality. To address these issues, we propose a pseudo-siamese Graph Neural Network method, namely MTINet+, which learns both biomedical network topological and molecule structural/chemical information as representations to predict potential interaction of given molecule and target pair. In MTINet+, 1-hop subgraphs of given molecule and target pair are extracted from known interaction of biomedical network as topological information, meanwhile the molecule structural and chemical attributes are processed as molecule information. MTINet+ learns these two types of information as embedding features for predicting the pair link. In the experiments of different molecule-target interaction tasks, MTINet+ significantly outperforms over the state-of-the-art baselines. In addition, in our designed network sparsity experiments , MTINet+ shows strong robustness against different sparse biomedical networks.

翻译：分子-靶点相互作用的研究对于药物发现中的靶点识别、先导化合物发现、通路研究、药物-药物相互作用等至关重要。现有方法大多利用生物医学网络信息或分子结构特征来预测潜在的相互作用关联。然而，基于生物医学网络信息的方法通常面临冷启动问题，而基于结构的方法由于结构/相互作用假设及数据质量限制，性能往往有限。为解决这些问题，我们提出一种伪孪生图神经网络方法——MTINet+，该方法同时学习生物医学网络拓扑信息和分子结构/化学特征作为表征，以预测给定分子-靶点对的潜在相互作用。在MTINet+中，从已知的生物医学网络相互作用中提取给定分子-靶点对的1跳子图作为拓扑信息，同时处理分子结构和化学属性作为分子信息。MTINet+通过学习这两类信息生成嵌入特征，用于预测分子-靶点对的关联。在不同分子-靶点相互作用任务的实验中，MTINet+显著优于现有最先进的基线方法。此外，在我们设计的网络稀疏性实验中，MTINet+对不同稀疏程度的生物医学网络表现出强鲁棒性。