For ML analysis, the best-fit model of our amino acid sequence datasets was estimated using the Akaike Information Criterion AICC) using ProtTest version 2.0 [ 67].
Ancestry of the selected HapMap datasets was estimated using different AIM panels (Table 1): Corach et al. [ 36] (COR), Galenter et al. [ 5] (GAL), Halder et al. [ 37] (HAL), Kosoy et al. [ 38] (KOS), Lao et al. [ 11] (LAO), Nassir et al. [ 39] (NAS), Phillips et al. [ 13] (PHI), and the commercial DNA Test Panel from Illumina (ILU; http://www.illumina.com/products/dna_test_panel.ilmn).ilmn
Pathogenicity predictions on variants identified within the analyzed dataset were estimated using MutPred [ 20], Polyphen-2 [ 21] and SNPs&GO [ 22] online software, producing 6 different scores and related prediction classes for each variant.
Phylogenetic relationships among the different members of the keratin gene family in the dataset were estimated using Bayesian and maximum likelihood approaches, as implemented in Mr. Bayes v3.1.2 [ 41] and Treefinder version March 2011 [ 42], respectively.
A best-fitting nucleotide substitution model for the dataset was estimated using the Akaike information criterion (AIC) as implemented in Modeltest v3.6 [47].
The phylogenetic tree for the dataset was estimated using Bayesian statistics implemented in MrBayes 3.1.2 [ 35].
The best-fitting model of amino acid substitution for each dataset was estimated using ProtTest v2.4 (Abascal et al. 2005) under the Akaike information criterion (starting tree: tree from a preliminary ML analysis using PhyML and a LG + Γ + I model with 8 rate categories for the γ distribution).
In the current study, when using the joint reference dataset, genomic predictions were estimated using a two-trait model, in which the same biological trait was considered to be a different trait in the CH and NH populations.
On this dataset, common fold changes were estimated using the MH method.
For each dataset, genes that are represented in less than 50% of the arrays were removed from this dataset, and missing values were estimated using KNNimpute with K = 10, Euclidean distance [ 77].
