Novel methods based on regression techniques to analyze multistate models and high-dimensional omics data.

Multistate models;Mann-Whitney U-test;Surrogate variables;Partial least squares;Backfitting;Batch effects

Sutirtha Chakraborty

University:University of Louisville

Department:Bioinformatics and Biostatistics

关键词: Multistate models;    Mann-Whitney U-test;    Surrogate variables;    Partial least squares;    Backfitting;    Batch effects;   

Others  :  https://ir.library.louisville.edu/cgi/viewcontent.cgi?article=1228&context=etd

美国|英语

来源: The Universite of Louisville's Institutional Repository

【 摘 要 】

The dissertation is based on four distinct research projects that are loosely interconnected by the common link of a regression framework. Chapter 1 provides an introductory outline of the problems addressed in the projects along with a detailed review of the previous works that have been done on them and a brief discussion on our newly developed methodologies. Chapter 2 describes the first project that is concerned with the identification of hidden subject-specific sources of heterogeneity in gene expression profiling analyses and adjusting for them by a technique based on Partial Least Squares (PLS) regression, in order to ensure a more accurate inference on the expression pattern of the genes over two different varieties of samples. Chapter 3 focuses on the development of an R package based on Project 1 and its performance evaluation with respect to other popular software dealing with differential gene expression analyses. Chapter 4 covers the thirdproject that proposes a non-parametric regression method for the estimation of stage occupation probabilities at different time points in a right-censored multistate model data, using an Inverse Probability of Censoring (IPCW) (Datta and Satten, 2001) based version of the backfitting principle (Hastie and Tibshirani, 1992). Chapter 5 describes the fourth project which deals with the testing for the equality of the residual distributions after adjusting for available covariate information fromthe right censored waiting times of two groups of subjects, by using an Inverse Probability of Censoring weighted (IPCW) version of the Mann-Whitney U test.