Methods for Integrative Analysis and Prediction Accounting for Subgroup Heterogeneity

Abstract

Multi-view data, where there are multiple data views (e.g., genomics, proteomics) measured on the same set of participants, have become increasingly available and require integrative analysis methods to fully utilize the available data and better understand complex diseases. At the same time, epidemiologic and genetic studies in many complex diseases suggest subgroup differences (e.g., by sex or race) in disease course and patient outcomes. While there are many existing methods to perform integrative analysis of multi-view data, we are unaware of any integrative analysis methods that also account for subgroup heterogeneity. Instead existing integrative analysis methods would require either (a) concatenating the subgroups which ignores any potential subgroup heterogeneity or (b) running a separate analysis for each subgroup which limits power especially in a high-dimensional data setting. While there are existing methods that account for subgroup heterogeneity, we are unaware of any that can also perform integrative analysis. These methods would require either (a) concatenating data views within each subgroup which fails to model the associations between data views or (b) considering each view separately which fails to fully utilize the multi-view data and requires combining results post hoc. This dissertation begins to fill this gap by proposing the novel statistical approach HIP (Heterogeneity in Integration and Prediction).Chapter 2 introduces HIP, a novel one-step method that (1) accounts for subgroup heterogeneity in multi-view data, (2) ranks variables based on importance, (3) can incorporate covariate adjustment, and (4) has efficient algorithms implemented in Python. The method introduced in this chapter can accommodate one or more continuous outcomes. Simulations show improved variable selection and prediction abilities compared to existing methods. We illustrate HIP using data from the COPDGene Study to identify molecular signatures that are common and specific to males and females and that contribute to the variation in COPD as measured by airway wall thickness. Chapter 3 extends HIP to accommodate multi-class, Poisson, and ZIP outcomes which allows researchers to study other clinically relevant outcomes. Simulations again show improved performance for HIP relative to existing methods in terms of variable selection and prediction abilities. We illustrate this method using data from the COPDGene Study to explore the genes and proteins associated with exacerbation frequency for males and females. One limitation researchers wishing to apply HIP to their data may encounter is that the implementation would require some knowledge of Python programming. Chapter 4 addresses this by introducing an R Shiny Application that provides a graphical user interface to the Python code allowing users to apply HIP to their own data. Users can select different analysis options or use the defaults provided. HIP, with improved accessibility through this R Shiny App, has many potential scientific applications.