Impact of gene expression data pre-processing on expression quantitative trait locus mapping

We evaluate the impact of three pre-processing methods for Affymetrix microarray data on expression quantitative trait locus (eQTL) mapping, using 14 CEPH Utah families (GAW Problem 1 data). Different sets of expression traits were chosen according to different selection criteria: expression level, variance, and heritability. For each gene, three expression phenotypes were obtained by different pre-processing methods. Each quantitative phenotype was then submitted to a whole-genome scan, using multipoint variance component LODs. Pre-processing methods were compared with respect to their linkage outcomes (number of linkage signals with LODs greater than 3, consistencies in the location of the trait-specific linkage signals, and type of cis/trans-regulating loci). Overall, we found little agreement between linkage results from the different pre-processing methods: most of the linkage signals were specific to one pre-processing method. However, agreement rates varied according to the criteria used to select the traits. For instance, these rates were higher in the set of the most heritable traits. On the other hand, the pre-processing method had little impact on the relative proportion of detected cis and trans-regulating loci. Interestingly, although the number of detected cis-regulating loci was relatively small, pre-processing methods agreed much better in this set of linkage signals than in the trans-regulating loci. Several potential factors explaining the discordance observed between the methods are discussed.