Network-Based Methods For Mining Pathogenesis Patterns In Cancer

The progress of medical research has enriched our understanding of cancer,but the morbidity and mortality of cancer in the world is still high,and there is an ascending trend year by year.This study aims to use the network-based model approach to excavate the pathogenic mode of cancer,and to provide information for cancer treatment.Medical network is a new approach by utilizing the method of network science,and it focuses on using network topology and network dynamics progression,identifying pathogenesis of diseases,finding drug targets,developing medical drugs and so on.Most of the studies on cancer pathogenesis detect the functional modules that cause cancer,lead to the progression and deterioration,on the basis of large-scale biomolecular data generated by biological experiments or biomolecular network data generated by mathematical modeling.One key issue in the current medial network research is to explore the messages encoded in the biomolecular network,reveal the network composition and its biological significance.Limited by the labor costs,material costs,time costs,and other objective factors,it seems to be an almost impossible task to verify the biological function in network modules by enumerating approach from the massive data.In order to effectively explore the functions of network modules,an increasing number of researchers begin to excavate the pathogenic module in biological network by using a combination of biological experiments and information calculation.This appraoach can detect the complex biological systems function modules easily and quickly,thus reducing the cost of the experiment.This study adopts the information calculation approach to excavate the pathogenic model that cause cancer,lead to the progression and deterioration,which is conducive to disease diagnosis,drug targets,drug repositioning and new drugs development.This paper mainly focuses on the algorithms based on network models targeting the network modules,cancer driver pathways,cancer dysregulated modules and cancer progression pathways.The main research and innovation are as follows.1.The present classical mining algorithm modules are mainly based on the idea of the dense inside and loosening outside.Since an object may participate in multiple functions,the fuzzy clustering models proposed in previous research are insufficient in dealing with network overlapping nodes.To solve this problem,this paper proposes a mining algorithm(Rough-Fuzzy Clustering,RFC)based on the unique advantages of rough sets and fuzzy sets in dealing with uncertainty and boundary problems.With RFC,the fuzzy equivalence relation is established by using fuzzy set,and the equivalence classes are obtained;the upper and lower approximation theory of rough set is used to determine whether a node is an overlapping node,and the overlapping functional modules are obtained.The comparison with other clustering algorithms shows that RFC is a simple and efficient algorithm,and that the modules obtained from RFC show high-level separation and significant physical meaning.2.A gene is often involved in a variety of biological functions,so the driving gene can belong to multiple cancer-causing driver pathways.This study presents a network-based model for overlapping driver pathway mining algorithm(Network-Based Method,NBM)through the construction of gene interaction networks.The new algorithm uses a greedy clustering strategy to add or delete nodes,which improves the computational efficiency and reduces the time complexity.NBM may directly detect driver pathways from somatic mutation data in cancer,without any prior knowledge.Driver pathways detected by NBM have higher enrichment,statistical significance and biological relevance.3.In order to integrate somatic mutation and gene expression data that are closely associated with cancer,establish a relationship between these two kinds of data effectively,and explore disorders modules in the progression of cancer,this study proposes a dysregulated module-mining method based on network models.The method constructs gene interaction network,and mines dense subgraph,that is dysregulated modules,by utilizing the widely existing mutually exclusivity between mutation genes in genome spectrum and the relationship between gene mutation and the gene expression variations.This model can be applied to the somatic mutation and gene expression data of any cancer to mine the dysregulated modules of cancer.In this study,the algorithm was applied to glioblastoma data,and the two dysregulated modules detected correspond to two subtypes of glioblastoma,and most of the genes in the two modules have been reported to be closely associated with glioblastoma;thus the algorithm has high prediction accuracy.4.Due to the time variations in detecting patients' cancer,clinically it is difficult to obtain the mutation data in multiple stages of the same patient's cancer progression.This paper proposes a network-based progression pathways-mining algorithm(Network-Based Inference,Net Inf)through constructing gene interaction networks with a large number of patient samples data.The algorithm mine cancer progression pathways from a large number of cancer patients,which is conducive to diagnose the stage of disease for clinicians,and to carry out corresponding therapy and drug targets.The method improves the computational efficiency and reduces the time complexity by constructing the mutually exclusive gene network.The comparison with other algorithms shows that the pathways detected by Net Inf have higher enrichment,thus providing new insights and approaches to studying the different stage of mutation gene set in cancer progression.