Study Of Biological Network Mining Based Function Information In Complex Disease

In the study of complex disease,the established network-based approaches can be mainly classified into three types,and these are:1)de novo inferring:these algorithms will deduce molecular network structures from omics data,then achieving the dysregulated relations among molecules in diseases.2)network comparison:these models will capture the molecular network differences across distinct pathological phenotypes,thus obtaining network biomarkers of biological or clinical importance.3)network module analysis:these methods will identify disease associated network modules or its functional activities,which indicate dysfunctional mechanisms for complex diseases.However,to the best knowledge,the existing approaches still have two major challenges to overcome:1)some can't fully exploit the functional annotations in biological networks,resulting in the identified genes or sub-networks lack of functional interpretations.2)some others considered the topological functionality,but the programming differences between distinct phenotypes were still largely unknown.To address these challenges,we proposed more effective algorithms to accurately quantify the phenotypic characteristics of functional networks,turning the researches of complex diseases from identifying topological associations into constructing functional associations.Our research works can be summarized as follows.Constructing network stratification analysis model?NetSA?based on the network flow balance.Different from the dense network structure by the traditional approaches,this model aims to extract sub-networks of great specificity for each biological function?i.e.,Gene Ontology terms?.To achieve the goal,the idea of balance flow was introduced to capture the functional subnetworks which are associated the complex disease.Moreover,we compared the NetSA model and the traditional approaches on the yeast and complex disease data sets,and validated the effectiveness of our model.Besides,the developed software could download at http://www.sysbio.ac.cn/cb/chenlab/images/NetSA_algorithm.rar.Constructing comparative network stratification analysis?CNS?based on the network flow balance and network feature selection model.CNS could capture the differential dysregulated networks under two disease states,then obtain the network biomarkers of functional interpretations.To achieve the goal,the concept of network flow balance was broadened to reach those gene nodes,which don't belong to but directly connect with the concerned functional gene set.And a relax restriction of network feature screening also was proposed to select the network biomarkers of functional interpretations.According to the results in various disease data sets,CNS presents accurate and reliable performance than the traditional identified biomarkers methods.And the free accessible website for software download is http://www.sysbio.ac.cn/cb/chenlab/images/CNSpackage₀.1.rar.Constructing differential function analysis model?DFA?based on joint non-negative matrix factorization model.In accordance with the intrinsic characteristics of biological data,a joint-NMF?joint non-negative matrix factorization?model of weak constraints is first proposed to derive functional network modules,in contrast to the joint-NMF models of strong constraints.Based on this model,we can identify the topological functional sub-networks as well as quantify the activities of these sub-network modules.The superior capabilities of the model are demonstrated in both numeric experiments and applications with real biological data sets.The source code package could download at http://www.sysbio.ac.cn/cb/chenlab/images/DFApack/age.zip.