Topological Approach to Analyses of Omics Data
Marija Đurđević · 2016
Genes expression is often a good indicator for prediction of patient's clinical results. In diseases such as cancer is inevitable to identify subcategories of phenotype. The goal of the Thesis is to use persistent homology on cancer tissue gene expression to identify new subgroups and try to predict the survival of patients in corresponding groups. We analyse the date from the International Consortium for Cancer Research. Simplicial complexes were built different resolutions using Vietoris-Rips algorithm. We counted the persistent homology and draw persistent diagrams. We developed a method for calculating confidence interval on persistent diagrams to precisely divide cancer subcategories. This method gave us promising results by discovering new subcategories and was accurate in prediction of patient clinical results. Results were obtained on data of different cancer types. Results were compared with different unsupervised learning methods.