Research On The Methodmof Identifying Cancer Driver Module Based On Network Model

The importance of cancer driver module to cancer precision medicine and personalized medicine makes the identification of cancer driver module become a research hotspot in bioinformatics.The research methods of this problem are mainly divided into two categories: one is the ab initio recognition method,the other is the recognition method based on prior knowledge.In this paper,the second type of method is mainly used to study the identification problem.Aiming at the characteristics of omics data such as excessive noise,incomplete and limited information of single omics data,multi-omics data information is integrated through protein interaction network to improve the integrity and accuracy of data,and a cancer-driven module identification method based on network model is proposed.The main work is as follows:Using somatic mutation,subcellular localization and protein interaction network data,we studied the recognition of common driver modules in R(R ?1)types of cancer.The module connectivity,mutual exclusion,coverage and the number of hops between genes within the module were taken as the module optimization objectives.A cancer common driver module identification model was proposed,and an identification method,IDM-SPS,was proposed to solve the model.In this method,subcellular localization data are used to denoise the protein interaction network,and the topology of the network is adjusted to reduce the possibility of negative impact of noise data on recognition.Somatic mutation data are used to weight the edge of the network,and the model is solved based on genetic algorithm containing five novel mutation operators.The algorithms IDM-SPS,Hotnet2 and MEXCOwalk are compared and analyzed on real biological datasets and simulated datasets.The experimental results show that the IDM-SPS method is better than the other two algorithms in the recognition of common driver modules in most cases.Using somatic mutation,gene expression and protein interaction network data,we studied the identification of specific driver modules of R(R(29)1)types of cancer.Somatic mutation data through the resumption of random walk algorithm is combined with gene expression data with protein interaction network of two weighted networks into two corresponding probability transfer matrix,add those two probability transfer matrix transfer matrix of the probability of a fusion,puts forward a model of cancer-specific driver module identification problem,Identified modules that differed between different cancer samples.Then,an extended algorithm ISM-SPG based on greedy strategy is proposed to solve the model.The DAMOKLE algorithm and ISM-SPG are compared and analyzed on real biological data sets.The experimental results showed that the specific modules identified by ISM-SPG method were better than those identified by DAMOKLE in most cases,with greater differences among different cancer samples.Moreover,compared with Damokle method,which can only identify differences in somatic cell data,the ISM-SPG method can identify modules with different gene expression levels among different cancer samples,which is conducive to better analysis of inter-cancer specificity.In summary,this paper studies the problem of cancer driver module recognition,and proposes two cancer driver module recognition models and algorithms based on network model.These methods may be useful complementary tools for cancer driver module detection.