A Multimodal Framework for Improving in Silico Drug Repositioning with the Prior Knowledge from Knowledge Graphs
Zhankun Xiong,Feng Huang,Ziyan Wang,Shichao Liu,Wen Zhang
DOI: https://doi.org/10.1109/tcbb.2021.3103595
2021-01-01
IEEE/ACM Transactions on Computational Biology and Bioinformatics
Abstract:Drug repositioning/repurposing is a very important approach towards identifying novel treatments for diseases in drug discovery. Recently, large-scale biological datasets are increasingly available for pharmaceutical research and promote the development of drug repositioning, but efficiently utilizing these datasets remains challenging. In this paper, we develop a novel multimodal framework, termed GraphPK (Graph-based Prior Knowledge) for improving in silico drug repositioning via using the prior knowledge from a drug knowledge graph. First, we construct a knowledge graph by integrating relevant bio-entities (drugs, diseases, etc.) and associations/interactions among them, and apply the knowledge graph embedding technique to extract prior knowledge of drugs and diseases. Moreover, we make use of the known drug-disease association, and obtain known association-based features from an association bipartite graph through graph embedding, and also take into account biological domain features, i.e., drug chemical structures and disease semantic similarity. Finally, we design a multimodal neural network to combine three types of features from the knowledge graph, the known associations and the biological domain, and build the prediction model for predicting drug-disease associations. Massive experiments show that our method outperforms other state-of-the-art methods in terms of most metrics, and the ablation analysis regarding the three types of features reveals that prior knowledge from knowledge graphs can not only lift the predictive power of in silico drug repositioning, but also enhance the model's robustness to different scenarios. The results of case studies offer support that GraphPK has the potential for actual use.