基于图卷积网络的药物与药物相互作用预测的研究进展

Research Progress on Predicting Interaction between Drugs Based on Graph Convolution Network

摘要: 药物之间的相互作用(DDIs)可能会导致严重的医疗伤害，确定DDIs并探讨其潜在机制对复合药安全至关重要。然而，在现实中由于需要对大量药物组合进行实验研究，分析检测DDIs仍然费时且昂贵。大多数药物相关知识是临床评估和上市后监测的结果，导致信息的数量有限。正确预测药物–药物相互作用不仅可以减少这些病例，还可以降低药物开发成本。因此，开发了许多计算方法来预测DDIs。其中基于图卷积网络的模型被提出用来预测DDIs。并在药物相互作用预测方面取得了最先进的结果。在此我们回顾了近些年基于图卷积网络的药物相互作用预测的研究现状，调查了近期研究人员在实验中利用到的最新方法和数据库。实验结果表明，基于图卷积网络的预测模型优于其他基于深度学习的算法。证明了基于图卷积网络来预测药物之间相互作用确实是有必要的和有价值的。

Abstract: Drug-drug interactions (DDIs) may lead to serious medical injuries. It is important to identify DDIs and investigate the underlying mechanisms of DDIs for the safety of compound drugs. However, in practice, analytical detection of DDIs is still time-consuming and expensive due to the need for experimental studies on a large number of drug combinations. Most drug-related knowledge is the result of clinical evaluation and post-marketing surveillance, resulting in a limited amount of information. Correctly predicting drug-drug interactions could not only reduce these cases but also reduce the cost of drug development. Therefore, a number of computational methods have been developed to predict DDIs. The model based on graph convolution network is proposed to predict DDIs and has achieved the most advanced results in drug interaction prediction. Here we review the research status of drug interaction prediction based on graph convolutional networks in recent years and investigate the latest methods and databases used by researchers in recent experiments. Experimental results show that the prediction model based on graph convolution network outperforms other algorithms based on deep learning. It is proven that it is necessary and valuable to predict drug-drug interactions based on graph convolutional networks.