Inference on gene expression change between two different samples is considered. We develop a mathematical model assuming that there exist two different functional states of a gene: “ON” and “OFF” . Each measured sample-specific gene expression intensity is described by an additive model, which accounts for fluctuations in absolute gene expression intensity and measurement error, to which a two-dimensional mixed normal model with four components considering the joint distribution of the sample “sum” and “difference” is approximated. We can successfully identify genes that are differentially expressed between two samples using posterior probabilities, while avoiding declaring false differences. The proposed methods are applicable to cDNA microarray data with two fluorescent dyes and to oligonucleotide data.
2005 The Biometric Society of Japan