Künstliche neuronale Netze in der genetischen Epidemiologie

Abstract

Gene-gene and gene-environment interactions play an important role in the etiological pathway of many complex diseases. However, common statistical methods like regression models have problems to capture the complex interplay between genetic and non-genetic factors. Artificial neural networks provide a great flexibility to model functional relationships and thus are a promising statistical tool to handle the complexity of biological interactions. The aim of this thesis is to explore the ability of neural networks to capture different structures of gene-gene and gene-environment interactions and to identify gene-gene interactions in simulation studies. In addition, the consistency of the estimated weights is investigated for non-identified neural networks. In summary, neural networks prove successful for exploratory analyses and particularly can be used if limited information on the kind of functional relationship between influencing factors and the investigated outcome is available.