When analysing large genetic datasets like UK Biobank, a common step in quality control is to ensure that the data you’re analyzing come from an ancestrally-homogenous group. Sophisticated software exists to estimate global ancestry proportions on a genome-wide scale. For quick, exploratory analyses, it is often useful to quickly identify individuals from a particular ancestral super-population.
superPop helps you to quickly guess which ancestral superpopulation someone is from. It’s designed for use with UK Biobank data.
It works by projecting individuals into principal component space using a common set of SNPs between 1,000 genomes and UKB. It helps you scale these PCs so they are comparable. It then applies a pre-trained random forest classifier to group each individual into their likely parent population.
This is meant for quick analysis, and clearly further results should be inspected in more detail for population structure.
You can install superPop from GitHub with:
# install.packages("devtools")
devtools::install_github("benjacobs123456/superPop")Before starting, please perform some basic SNP quality control on your data, and combine your SNP genotype data into a genome-wide PLINK1 binary fileset.
First, the function project_pcs() projects UKB individuals into PC space (a wrapper for plink2’s –score).
library(superPop)
project_pcs(bfile = "ukb_merged_chroms",path_to_plink = "plink.exe")
# This will produce an output file with the projected PC values called "projected_pcs.sscore"Second, we scale the PCs so they are comparable to those calculated with 1kg.
new_pcs = transform_pcs("projected_pcs.sscore")
# This will produce an output file with the projected PC values called "projected_pcs.sscore"Then we can predict which ancestral cluster each individual comes from using:
predict_ancestry(new_pcs)
# which will produce an output file, ancestry_estimates.tsv, containing estimates for each individual. You can plot these projections, grouped by predicted cluster, with:
plot_predicted_ancestry(projected_pcs = new_pcs, projected_ancestry = "ancestry_estimates.tsv")Which will give you a plot like this:
