A new approach for merging gene expression datasets

We propose a new approach for merging gene expression data originating from independent microarray experiments. The proposed approach is based upon a model assuming dataset-independent gene expression distribution, and dataset-dependent observation noise and nonlinear observation functions. The estimation algorithm combines smoothing spline estimation for the observation functions with an iterative method for gene expression estimation. The approach is illustrated by numerical results on simulation studies and real data originating from prostate cancer datasets.