Large scale prediction of phenotipic variables using gene expression data

Abstract

Motivation: Microarray technologies are increasingly being used in early drug development. When the gene expression data from microarray experiments also contains the IC50 values of a drug, it is of interest to predict the dosage based on the gene expression profiles. In this study, the supervised principal component analysis (SPCA) by Bair et al., (2006) was used to select and evaluate possible biomarkers and joint biomarker for the IC50 values of compound 352. Results: The response of interest is a vector of IC50 values of compound 352 the corresponding genes expression contains 7722 genes and 32 samples. This expression matrix was separated to genes negatively correlated to the IC50 and those positively correlated. The approach used in the SPCA method for biomarker (gene) selection is to select genes only the top k genes that are associated to the IC50, using this genes construct a joint biomarker and assess statistical significance for prediction using the joint biomarker.