Date of Award
Summer 7-25-2019
Degree Type
Dissertation
Degree Name
Doctor of Philosophy in Analytic and Data Science
Department
Statistics and Analytical Sciences
Committee Chair/First Advisor
Dr. Mingon Kang
Committee Member
Dr. Donghyun (David) Kim
Committee Member
Dr. Xuelei (Sherry) Ni
Committee Member
Dr. Herman (Gene) Ray
Committee Member
Dr. Jung Hun Oh
Abstract
Identifying complex biological processes associated to patients' survival time at the cellular and molecular level is critical not only for developing new treatments for patients but also for accurate survival prediction. However, highly nonlinear and high-dimension, low-sample size (HDLSS) data cause computational challenges in survival analysis. We developed a novel family of pathway-based, sparse deep neural networks (PASNet) for cancer survival analysis. PASNet family is a biologically interpretable neural network model where nodes in the network correspond to specific genes and pathways, while capturing nonlinear and hierarchical effects of biological pathways associated with certain clinical outcomes. Furthermore, integration of heterogeneous types of biological data from biospecimen holds promise of improving survival prediction and personalized therapies in cancer. Specifically, the integration of genomic data and histopathological images enhances survival predictions and personalized treatments in cancer study, while providing an in-depth understanding of genetic mechanisms and phenotypic patterns of cancer. Two proposed models will be introduced for integrating multi-omics data and pathological images, respectively. Each model in PASNet family was evaluated by comparing the performance of current cutting-edge models with The Cancer Genome Atlas (TCGA) cancer data. In the extensive experiments, PASNet family outperformed the benchmarking methods, and the outstanding performance was statistically assessed. More importantly, PASNet family showed the capability to interpret a multi-layered biological system. A number of biological literature in GBM supported the biological interpretation of the proposed models. The open-source software of PASNet family in PyTorch is publicly available at https://github.com/DataX-JieHao/