You are here:
PRACTICAL STRATEGIES FOR PROCESSING AND ANALYZING SPOTTED OLIGONUCLEOTIDE MICROARRAY DATA
Citation:
Bao, W, J E. Schmid, A K. Goetz, H. Ren, AND D J. Dix. PRACTICAL STRATEGIES FOR PROCESSING AND ANALYZING SPOTTED OLIGONUCLEOTIDE MICROARRAY DATA. Presented at Microarray Data Analysis Conference, Baltimore, MD, September 21 - 23, 2003.
Description:
Thoughtful data analysis is as important as experimental design, biological sample quality, and appropriate experimental procedures for making microarrays a useful supplement to traditional toxicology. In the present study, spotted oligonucleotide microarrays were used to profile gene expression in livers from animals exposed to conazole fungicides. Total RNA was extracted, reverse-transcribed and labeled with Cy dyes. Individual liver RNAs were labeled with Cy5. A reference RNA mixture was prepared containing equal amounts of each sample RNA and labeled with Cy3. Microarray processing was carried out in four experimental blocks, with each block containing one slide from each of the six treatment or control groups. The mouse microarrays contain 16,485 gene-specific spotted oligonucleotides, 192 negative control spots, and 773 empty spots or locations. Data processing and analysis strategies were developed to take advantage of the characteristics of these microarrays and data. First, the mean and median signal intensity and background values were transformed to the Logarithm base 2 scale. Second, data distributions were examined to select intensity measures (i.e., Log(median) vs. Log(mean)) that were consistent and normally distributed. Data distributions of Log(ratio) of cy5/cy3, with and without background subtraction, were also compared to decide if subtraction of local background was advisable for this dataset. Third, in order to define which genes were detectable, ratios between spot intensity and local background (I/B) of oligonucleotide spots were compared with I/B of empty locations and negative control spots within each microarray. This comparison was used to select a cutoff ratio for each microarray to distinguish "present" (i.e., detected) spots from those "absent" spots based on the data quality of each microarray. This cutoff ratio thus reflected the local background intensity and nonspecific binding for each array. "Present" genes were selected according to the frequency of "present" spots in each of the treatment and control groups. Only the genes defined to be "present" were analyzed further. An intensity-dependent local regression procedure (using SAS Proc Loess) was found to satisfactorily normalize the data. Following normalization, analysis of variance was used to look for differential expression between treated and control groups for each "present" gene (adjusting for experimental block variation). This presentation provides practical strategies for processing microarray data and demonstrates the importance of matching analysis methods to the characteristics of a dataset.