Research Of Prognostic Carcinoma Molecular Subtypes Based On Omics Data

The rapid development of science and technology has brought about the increasing progress of biological information technology.The rapid growth of biological data has triggered a data revolution.Over the years,a large number of different types of cancer data have been accumulated.The definition of cancer at the molecular level has always been an important research content in the field of bioinformatics.It is an important way to assist medical diagnosis and quickly provide diagnostic solutions for patients of different subtypes.How to quickly identify effective and diverse biological data Information has become one of the research hotspots in cancer molecular subtyping.At present,the common method of cancer subtyping is still semi-supervised or unsupervised sample typing for single omics data.Such methods cannot efficiently use a variety of data types related to cancer occurrence and development mechanisms,which may be able to cause information loss.Integrating multi-omics data can not only explore the relationship between cancer and related omics molecular data,but also discover the synergy between cancer data of each omics.Because biological data generally has the characteristics of wide feature dimensions,high noise interference,and small sample size,this thesis integrates multi-omics cancer data based on the method of similarity network fusion,and proposes to address the limitations of the original method using Euclidean distance,a similarity network fusion algorithm based on deep subspace model.For the fused similarity network,spectral clustering is used to achieve cancer subtyping tasks related to cancer prognosis.Because traditional methods of cancer subtyping algorithms mostly use continuous omics data,they ignore the significance of many discrete omics data for cancer subtyping.In this thesis,using the network propagation model,combined with the gene interaction network,the discrete and sparse somatic mutation data is smoothed by the network,thereby expanding the application range of the proposed fusion algorithm.In order to verify the similarity network fusion algorithm based on the deep subspace model proposed in this thesis,three types of omics data including m RNA,mi RNA and DNA methylation of five cancer data in TCGA(The Cancer Genome Atlas)database were used to verify the algorithm.We use the silhouette coefficient and log-rank p value to evaluate the final results.We also introduce somatic mutation data of three types of cancer data to verify the effectiveness of the use of network propagation models to smooth sparse discrete data to expand the scope of fusion algorithms.Finally,it is confirmed that the method proposed in this thesis can provide a certain reference for the cancer molecular subtyping task,prognostic diagnosis and treatment of cancer.