Masaru TAKEYA, Takehiro MATSUDA¹, Masao IWAMOTO², Norimichi TSUMURA¹, Toshiya NAKAGUCHI¹, and Yoichi MIYAKE¹

Division of Genome and Biodiversity Research, National Institute of Agrobiological Sciences, 2-1-2 Kannondai, Tsukuba, Ibaraki 305-8602, Japan
¹Graduate School of Science and Technology, Chiba University, 1-33 Yayoi-cho, Inage-ku, Chiba 263-8522, Japan
²Division of Plant Sciences, National Institute of Agrobiological Sciences, 2-1-2 Kannondai, Tsukuba, Ibaraki 305-8602, Japan

(Received September 5, 2005; Accepted December 6, 2006)

We propose a technique for estimating gene expression values for duplicated data on cDNA microarrays. In the scatter plots, the distribution is constructed from a mixture of normal two-dimensional distributions, which represent fluctuations in gene expression values due to noise. An expectation-maximization (EM) algorithm is used for estimating the modeling parameters. The probability that duplicated data is shifted by noise is calculated using Bayesian estimation. Six data sets of rice cDNA microarray assays were used to test the proposed technique. Genes in the data sets were subjected to clustering based on probability of true value. Clustering successfully identified candidate genes regulated by circadian rhythms in rice.

Key words: cDNA microarray, mixture distribution model, duplicated data, circadian rhythms