Integration Strategies of Cross-Platform Microarray Data Sets in Multiclass Classification Problem

Despite the increasing amount of available gene expression data, integrative analysis is still hindered by its high susceptibility to microenvironment fluctuations, resulting in inter-experiment variability known as batch effects. Therefore the development of data integration strategy is now more necessary than ever. Although several normalization algorithms have already been proposed, we believe that an effective model must rely on data migration between schemes. In this paper we apply this approach to a set of microarray data from core needle biopsy of breast cancers spanning different microarray platforms, and demonstrate its effectiveness in data preparation for unsupervised analysis and multiclass classification tasks. We propose a custom tool dedicated to defining the model structure. Additionally, we compare several pipelines of data processing, combining data normalization with different batch effect correction methods.