Print page

Artificial neural networks for modeling gene-gene- and gene-environment-interactions in genetic epidemiology

Description

The investigation of genetic factors is gaining importance in epidemiology. Most relevant from a public health perspective are complex diseases that are characterised by complex pathways involving gene-gene- and gene-environment-interactions. The identification of such pathways requires sophisticated statistical methods that are still in their infancy. Due to their ability in describing complex association structures, directed graphs may represent a suitable means for modelling complex causal pathways. This and a related project investigated the appropriateness for using directed graphs for modelling complex pathways in association studies.

The project investigated the ability of artificial neural networks to model and to identify gene-gene and gene-environment interactions. It showed that neural networks are very well suited to represent different interactions including discrete as well as continuous variables. In a second step, local and global model fit criteria were investigated. For interpreting the parameters, different approaches were studied like e.g. edge reduction systems and the concept of generalized synaptic weights. Within the project, the R-package neuralnet was implemented and published within the Comprehensive R Archive Network (CRAN).

Funding period

Begin:   August 2006
End:   July 2009

Contact

Frauke Günther

Sponsor

  • Deutsche Forschungsgemeinschaft (DFG)

Selected project-related publications