We fitted models in a step wise fashion starting with the variable with the strongest association, and keeping only variables with an effect on the outcome or confounding other factors in the final model.
Using backward elimination a final multivariate model was created keeping only variables with p < 0.05.
The resulting subset was further trimmed by keeping only structures with <40% sequence identity at the pairwise level.
Mapping was done with BWA [ 73] aligning pairs of reads and allowing maximum insert size to 500 bp and keeping only reads with mapping quality higher than 20.
A second round of selection of probe sets was performed by keeping only those with a minimum fold change of at least 4 times between these 2 conditions.
This included the Kaiser criterion of keeping only factors with eigenvalues greater than one, graphing and visual analysis of the scree plot of eigenvalues, and parallel analysis comparison of factor eigenvalues with Monte Carlo procedure randomly generated eigenvalues.
This included the Kaiser criterion of keeping only components with eigenvalues above 1, graphing and visual analysis of the scree plot of eigenvalues, and parallel analysis comparison of component eigenvalues with Monte Carlo procedures randomly generated eigenvalues.
Removal of genes expressed at very low levels (keeping only genes with counts per million (CPM) per sample > 1 in at least 8 samples) resulted in a set of 10,872 genes expressed in porcine peripheral blood.
We performed additional screening to reduce false positives, keeping only candidates with: 1) quality score ≥20; 2) sequence depth >20; 3) the approximate copy number of flanking sequences <2; 4) at least 1 uniquely mapped read for each allele; and 5) a minimum distance between SNPs ≥5 bp.
Gene and miRNA raw data were log2-transformed, normalized with Robust Spline Normalization and filtered, keeping only the probes with a detection P < 0.01 in at least one sample.
