Data Integration And Key Module Recognition Of High Dimensional Biomics Based On Random Nonnegative Matrix Decomposition

Recent technologies have enabled multi-platform genomic analysis of biological samples,such as DNA methylation(DM)and gene expression(GE),to produce socalled "multidimensional genomic data" simultaneously.These data provide a unique opportunity to study the coordination of multi-level regulatory mechanisms.However,there is a lack of comprehensive analysis of multi-dimensional genomic data for discovering combinatorial patterns.Non-negative matrix factorization algorithm was first proposed by Lee in 1999.After years of development,this method has become a mature algorithm and has been widely used in many fields.We hope to apply this method to biomics data.However,in actual calculation,the method has some problems,such as slow calculation speed and non-unique results.In this paper,we improve the algorithm by means of random projection and regularization terms,and obtain a random nonnegative matrix factorization algorithm.In this paper,SNMF algorithm is applied to four synthetic datasets and three real data-TCGA data.In the simulation experiment,the results of the four simulation data show that our algorithm has excellent performance in all aspects.Compared with the traditional non-negative matrix decomposition method,the proposed method has the advantages of fast calculation speed and high robustness of the results,which illustrates the superiority of the RNMF algorithm over the NMF algorithm.In the experiment of actual data,we use GO and KEGG pathway analysis to illustrate that the genes in the RNMF algorithm module are related to hepatocellular carcinoma.Finally,we found significant differences in survival rates between patients with positive tumor markers and patients with negative tumor markers.