Statistical applications in nutrigenomics : analyzing multiple genes and proteins in relation to complex diseases in humans

Background
The recent advances in technology provide the possibility to obtain large genomic datasets that contain information on large numbers of variables, while the sample sizes are moderate to small. This has lead to statistical challenges in the analysis of multiple genes and proteins in relation to complex diseases. In this thesis approaches are investigated to analyze large genomic datasets, taking complex relationships between genes, proteins and complex diseases into account. These approaches are applied to real data to investigate whether biologically relevant information from the dataset could be obtained or whether models could be obtained that are useful for diagnostic or prognostic purposes.

Results
We developed a general framework for the analysis of genetic, transcriptomic and proteomic data to obtain insight in biological mechanisms. This framework consists of the following steps: detection of heterogeneity, dimensionality reduction to deal with the large numbers of variables, statistical interpretation and biological interpretation. We found that within this multi-step approach application of a combination of methods, including methods that take interactions into account, is useful within the dimensionality reduction step. In this way more information is captured compared to applying only one method. After selection of relevant variables in the dimensionality reduction step, applying visualization tools, e.g. the interaction entropy graph, together with traditional statistical methods showed to be helpful for statistical interpretation whether variables contribute by their main and/or interaction effect to the outcome of interest. In the last step, biological interpretation of the statistical results was facilitated by literature search, pathway analysis and database mining.

Discussion
The general framework discussed in this thesis provides the possibility to analyze large nutrigenomic datasets. Although the contribution of genomic research to public health is at the moment limited, new advances in genomic research, e.g. genome-wide association studies, statistical approaches as discussed in this thesis, are promising and genomic research might in the near future lead to applications that translate into improvement of public health.