Impact of Molecular Biomarkers on Survival Outcomes: Signal Detection, Model Selection, and Statistical Inference in High-dimensional Settings

点击翻译按钮获取中文摘要

The goal of this project is to provide a theoretically justifiable and computationally feasible framework for analyzing large-scale data collected from epidemiology and other biomedical sciences, social science, marketing, environmental science, and econometrics. The proposed methods will provid a convenient means to identify the gene-specific impacts on survival and quantify the uncertainty of the estimates in high-dimensional settings. As a result, the methods will help detect patients' specific characteristics that make them respond differently to treatment or more susceptible to diseases, which is key to precision medicine. The PI further plans to disseminate the proposed research in education by redesigning graduate-level courses and training graduate students. The project, through its three major aims, features a series of methods to address fundamental issues arising from high-dimensional survival data analysis. The first aim is to detect gene-specific impacts on cancer patients' survival and to provide an integrated framework of detecting individual genes' relevance to survival when genes have heterogeneous patterns of influence. The second aim focuses on sequentially selecting important predictors for predicting survival outcomes. This approach is different from existing forward regression approaches, which are not applicable to handle censored outcome data with high-dimensional covariates. The third aim is to develop methods to quantify the uncertainty of survival models with high-dimensional predictors by integrating model selection, estimation, and inference steps.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.