| Alzheimer’s disease(Alzheimer’s Disease,AD)causes gradual loss of independent survival skills such as language ability and memory ability.In addition,due to the slow onset process,AD causes a heavy burden on individuals and society.However,humans still cannot fully understand the pathogenesis of AD,and there is no effective and safe ways for the diagnosis and treatment of AD.With the rapid development of sequencing technology,multi-omics data is growing exponentially,which provides opportunities and challenges for computational methods to identify AD-related omics features efficiently with low cost.Aiming at the identification of AD-related omics features,the study of multi-omics data-related methods will assist medical staff to improve diagnosis efficiency,reduce treatment costs,and provide opportunities for in-depth understanding of pathogenic mechanisms.Based on the central dogma of molecular biology,we researched computational methods to identify AD-related genes,AD-related small non-coding RNA molecule(MicroRNA,miRNA)which affects gene expression,AD-related drug targets which are the production of gene expression,and AD-related metabolites which are the final production of the genetic material.Through the study of computational methods,we explored the potential role and correlation of these AD-related omics features in AD,so as to reveal the pathogenesis of AD,and provide important support for clinical diagnosis and treatment methods.The main contents and contributions of this thesis are as following:(1)We studied identification method of AD-related genes based on multi-omics data.By merging the effect sizes of mutation sites in multiple genome-wide association analysis(Genome-Wide Association Studies,GWAS)data,and integrating GWAS data with multiple transcriptome datasets,a multi-omics data integration algorithm was developed for identifying AD-related genes,which can reveal the genetic pathogenic factors of AD by fully using multiple GWAS datasets and transcriptome datasets.In order to verify the effectiveness of our method,the identified AD-related genes are demonstrated by biological experiments in the literatures.The experimental results showed that the method can effectively identify the effect of gene expression differences caused by mutation sites on AD.(2)We studied identification method of AD-related miRNAs based on transcriptomics data.By constructing a gene-miRNA interaction network and classifying miRNA features,a semi-supervised classification algorithm was proposed for identifying AD-related miRNAs,which can extract low-dimensional features of network associations and circumvent the difficulty of obtaining negative samples and achieve high accuracy in identifying AD-related miRNAs.Compared with a variety of similar algorithms,the experimental results show that our method can effectively use the interactions between miRNAs and genes to identify the regulatory role of miRNAs in AD with best accuracy.The actual cases published in the literatures also verified the correctness of our identification results.(3)We studied identification method of AD-related interactions between drugs and targets based on proteomics data.A drug-protein pair network was constructed by drug-drug interactions,protein-protein interactions,and known interactions between drugs and proteins.A graph deep learning algorithm was developed for identifying AD-related interactions between drugs and targets,which can extract the features of the drug and protein in the drug-protein pair and the association between different drug-protein pairs.Compared with the six existing methods by two independent datasets,the experimental results show that our method has the best accuracy,which indicates building a drug-protein pair network can effectively help identify the interaction pattern between drugs and proteins.In addition,our method can also be used to identify the types of drug target interactions with high accuracy.In order to further verify our identification results,the potential targets and drugs of AD were discussed in detail,and the identification results were verified by cases published in the literatures.(4)We studied identification method of AD-related metabolites based on metabolomics data.A network association probability inference algorithm was proposed for the identifying AD-related metabolites through the semantic features of diseases and disease-related genes,which can infer metabolite similarity from disease similarity,and derive the probability matrix of AD-related metabolites.Comparing our method with other similar methods,the experimental results showed that our method improved the identification accuracy by traversing metabolites similarity network though effectively using disease semantic information and gene function similarity.The accuracy of the identified AD-related metabolites was verified through third-party biological experiments. |
