Date of Submission
Summer 7-31-2018
Degree Type
Thesis
Degree Name
Master of Science in Computer Science (MSCS)
Department
Computer Science
Committee Chair/First Advisor
Dr. Mingon Kang
Track
Others
Machine Learning
Chair
Dr. Mingon Kang
Committee Member
Dr. Donghyun Kim
Committee Member
Dr. Chih-Cheng Hung
Abstract
Glioblastoma multiforme (GBM) is the most fatal malignant type of brain tumor with a very poor prognosis with a median survival of around one year. Numerous studies have reported tumor subtypes that consider different characteristics on individual patients, which may play important roles in determining the survival rates in GBM. In this study, we present a pathway-based clustering method using Restricted Boltzmann Machine (RBM), called R-PathCluster, for identifying unknown subtypes with pathway markers of gene expressions. In order to assess the performance of R-PathCluster, we conducted experiments with several clustering methods such as k-means, hierarchical clustering, and RBM models with different input data. R-PathCluster showed the best performance in clustering longterm and short-term survivals, although its clustering score was not the highest among them in experiments. R-PathCluster provides a solution to interpret the model in biological sense, since it takes pathway markers that represent biological process of pathways. We discussed that our findings from R-PathCluster are supported by many biological literatures. Keywords. Glioblastoma multiforme, tumor subtypes, clustering, Restricted Boltzmann Machine