Data Adaptive Pathway Testing

Merrill D. Birkner, Division of Biostatistics, School of Public Health, University of California, Berkeley
Alan E. Hubbard, Division of Biostatistics, School of Public Health, University of California, Berkeley
Mark J. van der Laan, Division of Biostatistics, School of Public Health, University of California, Berkeley

Abstract

A majority of diseases are caused by a combination of factors, for example, composite genetic mutation profiles have been found in many cases to predict a deleterious outcome. There are several statistical techniques that have been used to analyze these types of biological data. This article implements a general strategy which uses data adaptive regression methods to build a specific pathway model, thus predicting a disease outcome by a combination of biological factors and assesses the significance of this model, or pathway, by using a permutation based null distribution. We also provide several simulation comparisons with other techniques. In addition, this method is applied in several different ways to an HIV-1 dataset in order to assess the potential biological pathways in the data.