For a global picture of differential gene expression, we used hierarchical clustering of the arrays. The differences between individuals were greater than the differences due to either ethanol treatment or phenotype: the ethanol treated and untreated samples from each person invariably clustered together, whether using all 31,522 probe sets expressed or the 5000 most variable probe sets (those with the largest coefficient of variation; data not shown). Although between-person effects were large, the paired design in which ethanol-treated and untreated LCLs from each of 42 individuals were used as repeated measures allowed us to detect the widespread effects of ethanol on gene expression, even when differences were small; each individual cell line acted as its own control, reducing the noise due to inter-individual differences.
