Research On Disease Module Identification Algorithm Based On Evolutionary Multi-objective Optimization

Complex diseases are caused by the dysfunction of biological system induced by the disease-associated genes,and their characters are affected by the interaction of multiple genes.At present,disease module mining and analysis based on biomolecular networks has become one of the most effective methods to reveal the mechanism of complex diseases.However,the network used for disease module identification is always constructed based on multi-samples.Thus,it is hard to make a disease module correspond to a single disease sample,which hinders the application of the achieved disease module for diagnosis.In this thesis,we first design an evolutionary multi-objective optimization algorithm based on the individual specific network to find the disease module that is closely connected and strong correlated with disease.Next,a disease module identification approach is proposed to detect disease module with good classification effects.The main research works of this thesis are as follows:(1)In this thesis,an evolutionary multi-objective optimal disease module identification algorithm based on individual specific networks is proposed(EMODMI).First,the EMODMI algorithm constructs individual-specific network based on single disease sample.Secondly,a multi-objective evolutionary algorithm is designed based on the individual specific network.The algorithm simultaneously optimizes the correlation strength between module and disease and internal tightness of the module,and designs a suitable initialization and population update guidance strategy for the disease module identification problem.Finally,the module score is proposed for evaluating the pros and cons of the disease module,and the final disease module is selected from a set of non-dominated solution sets obtained by the multi-objective evolutionary algorithm according to the module score.In the experimental phase,the performance advantages of EMODMI algorithm are analyzed by comparing with the four traditional disease module identification algorithm on two asthma data sets,and the biological significance of the identified disease modules is verified.(2)A classifiable disease module identification algorithm based on evolutionary multi-objective optimization(EMOCDMI)has been proposed to mine the disease modulethat have features such as close connection and strong correlation with diseases,as well as convenient to distinguish diseases from normal samples.First,the EMOCDMI algorithm inherits the network construction strategy of(1)to construct an individual-specific network for each disease sample.Secondly,a multi-objective evolutionary algorithm is designed to mine disease module.The difference between the algorithm and EMODMI is that the classification error rate is added as the optimization objective,and designs an interaction-based classification feature construction strategy based on individual-specific networks for objective function evaluation.The random crossover operator and the repairable mutation operator are also proposed to avoid the algorithm falling into the local optimum.Finally,according to the classification error rate,the optimal disease module was selected from the non-dominated solutions.Compared with the experimental results of a classifiable disease module identification algorithm and four disease diagnosis algorithms on eight datasets shows that the EMOCDMI algorithm can effectively identify the disease module that can distinguish disease from normal samples,and the effect of disease diagnosis based on module information is better than that based on molecule.