Universität zu Köln

Stephan, Johannes (2015). Understanding complex traits by non-linear mixed models. PhD thesis, Universität zu Köln.

Preview

PDF master.pdf - Accepted Version Bereitstellung unter der CC-Lizenz: Creative Commons Attribution. Download (1MB)

Abstract

Population structure and other nuisance factors represent a major challenge for the analysis of genomic data. Recent advances in statistical genetics have lead to a new generation of methods for quantitative trait mapping that also account for spurious correlation as caused by population structure. In particular, linear mixed models (LMMs) gained considerable attention as they enable easy black box-like control for population structure in a wide range of genetic designs and analysis settings. The aim of this work is to transfer the advantages of LMMs into a random bagging framework in order to simultaneously address a second pressing challenge: the recovery of complex non-linear genetic effects. Existing methods that allow for identifying such relationships like epistasis typically do not provide any robust and interpretable means to control for population structure and other confounding effects. The method we present here is based on random forests, a bagged variant of the well established decision trees. We show that the proposed method greatly improves over existing methods not only in identifying causal genetic markers but also in the prediction of held out phenotypic data.