基于二分图评价模型的网络药物靶标预测改进方法

Prediction of network drug target based on improved model of bipartite graph valuation

网络药理学作为新药研发领域中新的发展方向,受到越来越多的学者关注,而基因组药物发现研究中的一个关键问题就是如何识别药物与靶标蛋白质间新的交互作用。本研究即希望根据已知交互作用建立模型预测新的交互作用,以达到发现新靶标的目的。作者针对前人提出的二分图建模方法中存在的不足,提出了一种新的有监督的基于二分图评价模型的融合算法,根据已知的药物-靶标交互作用构建二分图网络,并建立药物-靶标蛋白质对的关联性评价模型,依此模型预测新的药物-靶标蛋白质交互作用,即预测新靶标。在已知交互作用数据集上做测试,本研究所提出的基于二分图评价模型的融合算法在性能上优于其他3种预测算法。基于二分图评价模型的融合算法集成化学空间、疗效空间和基因空间,构建药物候选化合物-靶标候选蛋白质交互网络,并建立交互作用预测模型,能预测出新的药物-靶标蛋白质交互作用,进而预测药物靶标,效果良好。

Network pharmacology,as a new developmental direction of drug discovery,was generating attention of more and more researchers.The key problem in drug discovery was how to identify the new interactions between drugs and target proteins.Prediction of new interaction was made to find potential targets based on the predicting model constructed by the known drug-protein interactions.According to the deficiencies of existing predicting algorithm based bipartite graph,a supervised learning integration method of bipartite graph was proposed in this paper.Firstly,the bipartite graph network was constructed based on the known interactions between drugs and target proteins.Secondly,the evaluation model for association between drugs and target proteins was created.Thirdly,the model was used to predict the new interactions between drugs and target proteins and confirm the new predicted targets.On the testing dataset,our method performed much better than three other predicting methods.The proposed method integrated chemical space,therapeutic space and genomic space,constructed the interaction network of drugs and target proteins,created the evaluation model and predicted the new interactions with good performance.