Compensation of scanner before robust M regression normalization in cDNA microarray
Hye Young Kim1, Jin Hyuk Kim2, Yong Sung Lee, Young Seek Lee, Tae Sung Park, Ki Woong Kim, and Hyun Ju Chang
1hykim121@hanyang.ac.kr, Hanyang Univeirsity College of Medicine; 2jhkim1@hanyang.ac.kr, Hanyang University College of Medicine
Normalization is one of the most important procedures to obtain accurate result of the experiment in cDNA microarray. In general, the normalization is applied on the data from image analysis, in a necessity of correcting the difference of amount of target RNA between control and experimental group, difference of labeling efficiency, difference of excitation and emission efficiency between Cy3 and Cy5, and difference of detector efficiency for them. These differences are the sources of error which affect to the gene expression ratio to be falsely obtained from the image analysis, so that a lot of normalization methods have been developed. However, the normalization methods developed by now are focused on correcting the gene expression ratio, not on the sources of the error.
The last two sources of error are from the scan process. Using microarray scanners, experimenters manually adjust the laser power for the excitation and emission of fluorescence and the gain of PMT for the detection of emitted fluorescence. The manual adjustment is necessary to obtain microarray images neither to be saturated nor to be dim. Different laser powers and gains of PMTs are to be applied to both fluorophores, because usually the fluorescence dye of Cy3 emits more bright fluorescence than the Cy5 when the laser is applied equally on both of them. There is no known relationship between the amount of fluorescence and the image intensity, but one can imagine that there is certain range shown linear relationship and the ranges out of it shown very low sensitivity between the amount of fluorescence and the image intensity. Therefore, we developed the governing equation of the image intensity as a sigmoid function of the amount of fluorescence, and what determine the sigmoid function are the parameters of the laser power and the gain of PMT.
For the application of the equation to the normalization, its reverse function was used to compensate the scanning process and robust M regression was followed. It makes it possible to go back to the pre-scanning state and analyze the microarray image to obtain the data closer to the gene expression ratio.