Analysis of gene expression data - Normalization of cDNA microarrays and large-scale response prediction

Abstract

In this dissertation we focus on the two areas in microarray experiments discussed in the previous section. Both areas are a part of the complete analysis of microarray data. The first area is situated in the early stages of the analysis, namely the preprocessing, while the second area is part of the final part of the analysis, the interpretation part. The first part of this dissertation is devoted to normalization of cDNA microarray data. We present a new family of normalization models based on linear mixed models and show how all normalization models can be expressed as a special case of a linear mixed model for normalization. We use two real datasets and a simulation study to illustrate the method and prove its validity. We show how the method can be extended to develop models with nonconstant variability. The first part of the dissertation ends with a chapter devoted to software, which discusses the development of a graphical user interface to facilitate the use of the mixed model for normalization. The second part is devoted to response prediction using microarray data. We propose a new method based on a weighted resampling scheme in combination with existing methods such as Lasso and Elastic Net. We use four real datasets, all with IC50 as response, to compare the new method to existing methods. Furthermore, two simulation studies are used to fine-tune the method and to compare its performance to existing methods. Again the part ends with a software section, which discusses the idea of parallel computing to implement the algorithm as efficient as possible.