To avoid analyzing genes that were not expressed, only probe sets that were called “present” in at least 33% of the arrays in at least one experimental group (phenotype, treatment, sex) were selected for analysis (McClintick and Edenberg, 2006). Using these criteria, 31,528 of the 54,675 probe sets on the GeneChips were retained for analysis. The MAS5 data were imported into Partek Genomics Suite (Partek Inc., St. Louis, Mo.). Because we expected cell lines from different individuals to differ, analysis was done using a general linear method with repeated measures for 0 and 75 mM ethanol, with the main effects factors ethanol treatment, phenotype (alcoholic vs. non-alcoholic), sex and labeling batch. Addition of the three interaction terms (sex*treatment, sex*phenotype, and phenotype*treatment) to the model did not improve the results; none of the interaction terms reached significance after correcting for multiple testing. Therefore, we present the data from the simpler model with main effects only. P-values for each factor tested were imported into R to compute false discovery rate (FDR) using the Storey q-value package (Storey and Tibshirani, 2003). Partek Genomics Suite was used for hierarchical clustering of the arrays using Euclidean distance and average linkage.
