Sijian Wang - University of Wisconsin - Madison
Main Content
Our work is motivated by analyzing TCGA ovarian cancer data with survival outcome. Our analysis has two aims. One is to identify important core pathways and important genes within the identified pathways related to ovarian cancer survival. The other is to build a predictive model for future patients' survival based on the identified genomic features. We propose two methods. The first method is doubly regularized Cox regression (DrCox), which is based on the penalized partial likelihood estimation with a mixture of convex penalties. The convexity of objective function makes the method numerically stable especially when the number of predictors far exceeds the number of the observations. A fast coordinate descent algorithm is exploited to avoid matrix operations and speed up the computation. This is joint work with Tongtong Wu from the University of Maryland. The second method is pathway-based index model. Motivated by the concept of personalized medicine, the proposed hierarchical framework models a survival related phenotype as attributable to known genes within known biological pathways. Given genes identified as conferring increased or decreased risk within a pathway, a pathway summary, or index, is then constructed. The indices are used in a second model to identify important pathways and predict future patients' survival. Using TCGA data, we show that the patient-specific index scores across important pathways (referred to as patient-specific risk profiles) are powerful and efficient characterizations useful in addressing a number of questions related to predicting survival and optimizing treatment. This is joint work with Kevin Eng and Christina Kendziorski from the University of Wisconsin, Madison. Contact Yu Zhang – 867-0780 for additional information