Reads aligned to known genes were analyzed by edgeR (Robinson, McCarthy, & Smyth, 2010). Counts for each sample were normalized to counts per million reads. To avoid analyzing genes that were not expressed or expressed at near background levels, only genes that had more than 2 counts per million reads in at least 20 samples were retained for analysis. Data were examined by Multidimensional Scaling in the edgeR package (Robinson et al., 2010) and principal components analysis in Partek Genomics Suite (version 6.6, Partek, Inc. St. Louis, Mo) to detect outliers. One outlier (female, alcoholic, untreated) was found (see Supplemental Figure 1), so we removed it and its treated pair from further processing. In addition, we discovered that the demographic info for one of the control samples (male) did not match the phenotype listed for the sample, so it was also removed from analysis, leaving 20 controls and 20 AD. Dispersion estimates (common, trend and tag dispersion) were calculated and used in the analysis by general linear methods in edgeR to identify differentially expressed genes.
