Fig 1. Flowchart showing the general methodological approach underpinning the pipeline.

In high dimensional genetic data of n samples with p genotyped SNPs, the number of SNPs was first reduced from p to k by means of the RF layer. The selected k SNPs were further reduced by means of two alternative methods, the ensemble of three regression methods and the Boruta method. The most significant SNPs (key SNPs) are those that were selected by majority of the methods, i.e. in consensus, during the final iteration.