Non-negative Matrix Factorization Model For Tumor Heterogeneity Studies

Cancer is a kind of disease caused by abnormal cell division and apoptosis.Tumor heterogeneity is one of the characteristics of malignant tumors,because of its inability to fully analyze its subclonal structure,coupled with easy to produce resistance and metastasis after treatment,the treatment of great challenges.DNA methylation is an epigenetic modification of DNA molecules and is believed to be the cause of cancer and other human diseases.It is helpful to understand the development process of tumor by estimating the proportion of each subclone of DNA methylation.In this thesis,we first introduce several methods for the estimation of tumor purity and differential methylation analysis.Secondly,the following computational models of tumor heterogeneity decomposition via Nonnegative Matrix Factorization and quadratic programming are established,which are based on the different types of molecular data in TCGA database.First,the tumor heterogeneity model based on DNA methylation;Second,tumor heterogeneity model based on gene expression;Third,tumor heterogeneity model of DNA methylation and gene expression.The simulation results show that our method can successfully estimate the proportion of each subpopulation within the tumor,and the model based on multi data integration has a higher accuracy than the single data type.As a special case,our method has a very good agreement with the InfiniumPurify,ABSOLUTE,ESTIMATE and other methods for the estimation of the tumor purity(the number of tumor subsets is 2).