Statistical analysis

Statistical analysis was performed with R software (www.R-project.org) and packages available through Bioconductor(www.bioconductor.org). Quality control was performed, as previously described.⁶³ Expression values were obtained by using the GC Robust Multiarray Average algorithm. Batch effects were removed by identifying and adjusting for surrogate variables with the sva package.⁷⁰ Probe sets with 15 or more samples and expression values of greater than 3 were kept for analysis. Data were log₂-transformed and fitted with a linear model. Fold changes (FCHs) between comparisons of interest were estimated, and hypothesis testing was conducted by using contrasts under the general framework for linear models in the R software limma package. P values from the moderated (paired) t test were adjusted for multiple hypotheses by using the Benjamini-Hochberg procedure. Genes with FCHs of greater than 2 and a false discovery rate (FDR) of less than 0.05 were considered differentially expressed. Hierarchical clustering of samples/conditions used a McQuitty agglomeration algorithm. Gene-set variance analysis was performed by using unsupervised sample-wise enrichment.⁷¹ Gene-set overrepresentation analysis was performed with XGR software⁷² and crowd-extracted expression of differential signature disease signatures.⁷³

All 29 samples from patients with ichthyoses, 21 control subjects, and 10 patients with psoriasis and AD were available for RT-PCR analyses. mRNA expression RT-PCR data were log₂-transformed and fitted to a linear model as above. Values of less than the limit of detection were imputed as 20% of the minimal value over the limit of detection, as previously eported.⁷⁴ See the Methods section in this article’s Online Repository for details.