Principal component analysis, k-means, and ADMIXTURE (dx.doi.org/10.17504/protocols.io.bkwbkxan in protocols.io)

The pipeline for running PCA, k-means, and ADMIXTURE from a single dataset can be downloaded: https://github.com/jbv2/VCF2PCP. In brief, from the IPVS, we kept only our NM samples and the 4 NP individuals from the 1000 genomes project (samples ids: HG01926, HG01938, HG01961, HG02272). We kept biallelic SNVs with a MAF > 0.05 with bcftools v1.9-220-gc65ba41, and removed variants in linkage disequilibrium (r2 > 0.85) with bcftools +prune plugin using parameters—window 2000bp—nsites-per-win 1. We transformed VCF files into Eigenstrat format. PCA was performed using Smartpca from Eigensoft v6.1.4 [72] requesting numoutevec: 20. We kept eigenvectors with P-value < 0.01, then recalculated the percentage of variability per eigenvector, being 100% the sum of the selected eigenvalues. In k-means analysis we calculated the Average Silhouette method to define optimal clustering. For ADMIXTURE v1.3 [20] analysis we used the—seed 43 parameter.