【作者】 张强；

【导师】 郑春厚；

【作者基本信息】 安徽大学 ， 计算机科学与技术， 2022， 硕士

【摘要】 传统的药物研发是一个长时间、高投资和高风险的过程。随着医学技术的发展,人们对疾病的发生机制有了越来越清晰地认识,对药物的需求也在不断增加。传统的药物研发方法已经很难满足人们的需求,人们需要一种新的药物研发方法。药物重定位是一种为已有药物寻找新的适应症的方法,可以显著地加快研发进程、减少研发费用和降低研发风险,受到了越来越多的关注。此外,随着计算机技术和高通量测序技术的发展,研究人员能够从多种来源获得多种类型的数据,这为药物重定位研究提供了前所未有的机遇和挑战。虽然研究人员已经提出了许多药物重定位的方法,但如何有效整合多种来源的数据仍然是药物重定位的一个难题。本文主要研究通过有效整合多源数据提高药物-疾病关联预测的准确性。主要工作如下:1.提出了一种基于网络一致性投影的药物-疾病关联预测方法。首先,基于药物化学结构相似性、药物靶标域相似性和药物靶标注释相似性计算一种新的疾病相似性。其次,利用疾病和药物的拓扑结构信息计算疾病和药物的高斯核相似性。然后,使用相似性网络融合算法分别融合多种药物相似性和疾病相似性。最后,基于融合后的药物相似性网络和疾病相似性网络以及药物-疾病关联网络,使用网络一致性投影方法预测新的药物-疾病关联。实验结果表明,该方法的预测性能优于几个已有的药物重定位方法,并且能够将其用于预测特定药物的新的关联疾病。2.提出了一种基于图自编码器的药物-疾病关联预测方法。首先,使用亲和网络融合方法将药物化学结构相似性、药物靶标域相似性和药物靶标注释相似性融合成一种药物相似性。然后,基于融合后的药物相似性网络、疾病语义相似性网络和药物-疾病关联网络构建异质网络。使用图自编码器提取药物和疾病的多层嵌入表示,并分别将多层嵌入表示串联起来作为药物和疾病最终的嵌入表示。最后,将药物和疾病的嵌入表示串联起来作为药物-疾病对的特征用于训练一个多层感知机模型以预测潜在的药物-疾病关联。实验结果表明,该方法的预测性能优于其它几个先进的药物重定位方法。更多还原

【Abstract】 Traditional drug development is a long,high-investment and high-risk process.With the development of medical technology,people have a more and more clear understanding of the mechanism of disease,the demand for drugs is also increasing.Traditional drug development methods have been difficult to meet the needs of people,so people need a new method of drug development.Drug repositioning,a method of finding new indications for existing drugs,has received increasing attention as it can significantly speed up the development process,reduce development costs and reduce development risks.In addition,advances in computer technology and high-throughput sequencing technology allow researchers to access multiple types of data from multiple sources,providing unprecedented opportunities and challenges for drug repositioning research.Although researchers have proposed many methods for drug repositioning,how to effectively integrate data from multiple sources is still a difficult problem for drug repositioning.This thesis focuses on improving the accuracy of drug-disease association prediction through effective integration of multi-source data.The main work is as follows:1.A drug-disease association prediction method based on network consistency projection was proposed.Firstly,a novel disease similarity was calculated based on drug chemical structure similarity,drug target domain similarity and drug target annotation similarity.Secondly,the Gaussian kernel similarity of diseases and the Gaussian kernel similarity of drugs were calculated by using topological information of diseases and drugs.Then,the similarity network fusion algorithm was used to fuse multiple drug similarities and disease similarities,respectively.Finally,based on the fused drug similarity network and disease similarity network as well as the drug-disease association network,the network consistency projection method was used to predict new drug-disease associations.The experimental results show that the proposed method outperforms several existing drug repositioning methods and can be used to predict new associated diseases for a specific drug.2.A drug-disease association prediction method based on graph auto-encoder was proposed.Firstly,the drug chemical structure similarity,drug target domain similarity and drug target annotation similarity were fused into a drug similarity by affinity network fusion method.Then,a heterogeneous network was constructed based on the fused drug similarity network,disease semantic similarity network and drug-disease association network.The multi-layer embedded representations of drugs and diseases were extracted using graph auto-encoder,and multi-layer embedded representations were connected respectively as the final embedded features of drugs and diseases.Finally,the embedded representations of drugs and diseases were concatenated as features of drug-disease pairs to train a multilayer perceptron model to predict potential drug-disease associations.The experimental results show that the prediction performance of this method is superior to several other advanced drug repositioning methods.更多还原