Biclustering Algorithm And Its Application To Breast Cancer Subtype Analysis

Breast cancer is a malignant tumor,which is a serious harm to women’s health.Its incidence increases and patients become younger year by year.In clinical,breast cancer shows significant heterogeneity,and single treatment is unable to meet the needs of different types of breast cancer patients.Recent research has discovered that the occurrence of breast cancer is the accumulation of many kinds of factors,among which abnormal gene expression is one of the most common features.As an important epigenetic mechanism,abnormal DNA methylation is very common in key genes correlated with breast cancer.DNA methylation plays a key role in the occurrence and development of breast cancer.Therefore,it is necessary to seek the target of breast cancer subtypes and the methylation pattern for the diagnosis,treatment and prognosis of breast cancer by analysing gene expression profiles and methylation level.Currently,most study of breast cancer subtypes put emphasis on traditional one-way clustering.However,the traditional clustering can only cluster data in one direction at a time,and it will lose a lot of local information.It also does not allow the overlap of two classes.As a matter of fact,genes are always involved in many biological pathways and play different roles.In this paper,aiming at solving the shortcomings of traditional clustering,we analyzed the gene expression data from the perspective of biclustering,and discussed the classification of breast cancer.The main work are concluded as follows:1.Considering the complexity and heterogeneity of the relationship between gene expression and breast cancer subtypes,this paper proposes a biclustering algorithm based on AP and ISA clustering.Firstly,the samples of breast cancer data set are clustered into different classes according to AP clustering.Randomly generated seeds of ISA algorithm are classified according to the result of AP clustering.Then ISA algorithm is performed.Finally,the resulting clusters of the breast cancer samples are obtained.Each cluster includes a subset of samples and a subset of genes,and biclusters’ overlap is allowed.In the clustering process,the AP clustering results add priori knowledge to the seed selection of the ISA algorithm,Thus the reliability of the results can be guaranteed.2.In this study,the breast cancer data set with the "intrinsic gene" was divided into different types.Firstly,we obtained the genes associated with each subtype.it is shown that there is a significant difference among the genetic components of each subtype.Secondly,by analysing the clinical data of breast cancer patients,we find that each subtype has its own characteristics in ER and PR.3.The experimental results showed that all the genes of breast cancer subtypes have good biological interpretation.First of all,the effects of genes in different subtypes on the occurrence and development of breast cancer were analyzed by searching the published papers.Secondly,by biological pathway enrichment analysis of all gene sets in each subtype,we found that these genes were significantly enriched in several key pathways of breast cancer,which proved that they play an important regulation role in the occurrence and development for breast cancer.Biological pathways of different subtypes exist significant differences.4.Methylation level of each breast cancer subtype is studied.First of all,the methylation loci in different subtypes were classified according to the position distribution.Then,in each subtype,the average value of methylation in the same region of each sample was calculated.The results showed that there were significant differences in methylation levels of each breast cancer subtype,and each subtype has its own methylation pattern.In summary,this study innovatively established different gene expression patterns and DNA methylation pattern of breast cancer subtypes.It can provide reference for the prognosis,diagnosis and treatment to breast cancer in new direction.