Network Based Methods For Disease-Related Patterns Mining

With the rapid development of science and technology,biomedical research based on network medicine systematically explores the molecular mechanism of complex diseases focusing on the research of the different patterns mining,which reveal the complex disease related biological markers and the drug target.At the same time,the rapid development and application of high-throughput technology has brought a wealth of multi-omics data,but also brought challenge about the integration of these heterogeneous data analysis.Because of the complexity of cancer,the understanding of its pathogenesis becomes difficult,which requires the use of biological information tools to analyze,explore and extract new biomedical knowledge from multiple sets of data.In this study,we try to combine multi-omics data with biological network to explore the pathogenesis of complex disease-related factors,and analyze the relevant model for the discovery and treatment of disease.We focus on the four aspects: disease genes,driver genes,dysregulated pathway and inferring network.The main work and innovation are as follows:1.For the problem of disease gene mining,we proposed a method to extract the pathogenic genes by combining the network feature and the biological characteristics.By defining the network motif score index with the change relation between differential methylation and different expression pattern,we identified coherent changed motifs and disease-related genes.The results of colorectal cancer dataset showed that the method was superior to the existing method in the classification performance.The identified network motif and the predicted gene were highly correlated with the occurrence of colorectal cancer and were enriched with cancer markers.2.For the problem of driver genes mining,we established a network by considering the relationship between gene mutation and gene expression,the relationship between mutated genes and heterogeneity of patients,and the algorithm was combined with network diffusion step and agglomeration sorting algorithm.By applying it to three cancer data sets(pleomorphic glioblastoma,ovarian cancer,breast cancer),our method is superior to the comparative methods in some evaluation indicators.At the same time,our method predicted 58,55,13 potential drivers,respectively,most of which were known to drive genes,and the rest of the genes were further confirmed to be associated with cancer by functional analysis and co-citation analysis.Our method can obtain better results,especially in the recognition of drive genes with low-frequency mutation.3.For the problem of mining and analyzing the pattern of dysregulated pathways,we proposed a method to identify the dysregulated pathway by an optimal strategy method integrating the multi-omics data.The method was applied to the data of breast invasive carcinoma(BRCA)including DNA methylation,DNA copy number,somatic mutations and gene expression profiles,and identified the disorder pathways of different disease subtype.The results showed that the first 30 dysregulated pathways appeared common and specific dysregulation patterns,and 44 differentially expressed genes in the genetic and epigenetic disorders of specific subtype pathways were identified.Literature validation and functional enrichment analysis further confirmed that they are highly correlated with BRCA.Our method provides a new way of integrating the multi-omics data into the identification of disorders and disease genes.4.For the problem of inferring regulation network of cell cycle genes,we proposed a method of integrating multi-platform transcriptional data,and using dynamic cascade method to construct the transcription factor regulated network of cell cycle genes in each stage of disease.The method was applied to the subtype of lung cancer-lung adenocarcinoma,and five stages of dynamic gene regulated network were constructed.The feasibility of network inferring about lung adenocarcinoma was confirmed by network enrichment analysis,functional analysis,literature verification and correlation analysis.