Analysis of Pair-Wise Comparisons

For biological arrays, performing multiple experimental repeats is a common way to improve certainty of gene identification. But this is offset by the high cost of each experiment. In a typical experiment, n independent measurements (e.g., data from n separate gene expression arrays (Affymetrix, 2002; Lipshutz et al., 1999) are combined to yield a single estimation per assay (e.g., per gene). Thus, 2n independent measurements, yields only n comparative measurements. We show that it is possible to compute abstractly the probabilities appropriate for using other subsets of the potential n comparisons derivable from such data. We show the 2n − 1 independent comparisons available from among the n possibilities perform the best.