Development of a Reproducibility Index for cDNA Microarray Experiments

Since its introduction in 1995 by Schena et al. cDNA microarrays have been established as a potential tool for high-throughput analysis which allows the global monitoring of expression levels for thousands of genes simultaneously. One of the characteristics of the cDNA microarray data is that there is inherent noise even after the removal of systematic effects in the experiment. Therefore, replication is crucial to the microarray experiment. The assessment of reproducibility among replicates, however, has drawn little attention. Reproducibility may be assessed with several different endpoints along the process of data reduction of the microarray data. We define the reproducibility to be the degree with which replicate arrays duplicate each other. The aim of this note is to develop a novel measure of reproducibility among replicates in the cDNA microarray experiment based on the unprocessed data. Suppose we have p genes and n replicates in a microarray experiment. We first develop a measure of reproducibility between two replicates and generalize this concept for a measure of reproducibility of one replicate against the remaining n-1 replicates. We used the rank of the outcome variable and employed the concept of a measure of tracking in the blood pressure literature. We applied the reproducibility measure to two sets of microarray experiments in which one experiment was performed in a more homogeneous environment, resulting in validation of this novel method. The operational interpretation of this measure is clearer than Pearson's correlation coefficient which might be used as a crude measure of reproducibility of two replicates.