Reproducibility of microarray studies: concordance of current analysis methods

Abstract

Publisher Summary Variation between analysis protocols is an important issue, as numerous tools have been developed and applied to the problem of measuring the global transcriptome using microarrays. Many of these tools differ radically in their theoretical underpinning. The chapter describes the microarray-data analysis pipeline making special reference to the methods appropriate to the example data set presented in the chapter. The sequential steps incorporated by the microarray studies in the data analysis pipeline are pre-processing of data, detection of differentially expressed genes, functional profiling of the lists of differentially expressed genes, and validation. Procedures for the assessment of data quality are Outlier removal and choice of disease marker. These procedures are applicable to many microarray studies of clinical samples. The performance comparison helps in the comparison of methods available at four key stages in the data analysis pipeline—(1) definition of probe sets, (2) computation of expression measures, (3) correlation analysis, and (4) FDR control. Transcriptomics studies of neuropsychiatric disorders are sensitive to the choice of analysis protocol because the gene expression changes observed are generally subtle. While the field of microarray data analysis is maturing, there is yet no consensus on optimal protocols. An alternative method of finding genes that show interesting patterns of expression across the sample set is the approach of cluster analysis. The chapter demonstrates discordance between the content of candidate gene lists output from different microarray data analysis protocols performed on the same microarray data set.