ancientpca

A Probabilistic Approach to Conducting PCA with Sparse Genotype Data from Ancient Samples

Install and load in R

library(devtools)
install_github('sojwolf/ancientpca')
library(ancientpca)

Dependencies

R packages you need installed:

library(vcfR)
library(softImpute)
library(ggplot2)
library(ggrepel)
library(cowplot)

Input data

1. Genotypes: vcf-file with genotypes only (i.e. "0/0", "0/1", etc.). Missing values can be coded as "." or "./."
2. Meta Data: csv-file that contains headers "Sample_ID" and "Population" (can contain other columns too) for at least all samples in your vcf-file.

Example workflow

1. Load vcf-file into R and subsample, if need be by setting the maximum allowed missingness per sample and SNP. The default is set to 1, i.e. 100%. The vcf-file can be zipped.

gt <- vcfToGenotypeMatrix(vcf_file="vcf_file.vcf", max_missing_snp=0.9, max_missing_sample=0.95)

2. Impute missing genotypes using package softImpute

gt_imputed <- impGenotypeMatrix(genotype_matrix=gt)

3. Calculate PCA using R's prcomp function

pca_gt <- prcomp(gt_imputed, center = TRUE, scale. = TRUE)

4. Plot PCs 1-6 and save output to PDF file.

plotImpPCA(pca_obj=pca_gt, imputed_matrix=gt_imputed, original_matrix=gt, max_missing_snp=0.9, max_missing_sample=0.95,
  meta_file = "meta.csv", output_pca_pdf = "output.pdf")