Genetic Ancestry Inference for Pharmacogenomics

Abstract

Genetic ancestry inference can be used to stratify patient cohorts and to model pharmacogenomic variation within and between populations. We provide a detailed guide to genetic ancestry inference using genome-wide genetic variant datasets, with an emphasis on two widely used techniques: principal components analysis (PCA) and ADMIXTURE analysis. PCA can be used for patient stratification and categorical ancestry inference, whereas ADMIXTURE is used to characterize genetic ancestry as a continuous variable. Visualization methods are critical for the interpretation of genetic ancestry inference methods, and we provide instructions for how the results of PCA and ADMIXTURE can be effectively visualized.

1. Introduction

Pharmacogenomic variants that mediate patients’ response to medications often show large allele frequency differences among population groups [1, 2]. These allele frequency differences have important implications for treatment decisions, with population-specific effects observed for drug efficacy, dosage, and toxicity. Indeed, there are numerous examples of racial and ethnic differences in drug response, many of which can be attributed to allele frequency differences in pharmacogenomic variants [3-7]. It has been observed that up to 20% of newly approved drugs show distinct racial and ethnic response profiles, and differences of this kind can lead to group-specific treatment recommendations issued by the FDA [8].

Nevertheless, it should be stressed that race and ethnicity are socially ascribed characteristics, based on shared origins, culture, heritage, and social experiences. Race and ethnicity are not biological categories and are therefore imprecise proxies for genetic diversity [9]. Genetic ancestry, on the other hand, is a characteristic of the genome. Genetic ancestry measures individuals’ biogeographical origins, based on correlated allele frequency differences among ancestral source populations [10]. Genetic ancestry can be defined independently of the social dimensions of race and ethnicity, and it can be characterized objectively and with precision, as either a categorical or a continuous variable. Accordingly, pharmacogenomic variation among populations is better modeled with genetic ancestry as opposed to race and ethnicity.

The aim of this chapter is to provide a practical guide to genetic ancestry inference for pharmacogenetic researchers who may wish to stratify their study cohorts based on patterns of genetic diversity rather than, or in addition to, the more commonly used social categories of race and ethnicity. In light of the increasing availability of large-sale genomic datasets, we focus on genetic ancestry inference methods that make use of genome-wide genetic variant data, including whole-genome sequences, whole exome sequences, and whole-genome genotypes. We provide detailed protocols for two commonly used methods – principal components analysis (PCA) and ADMIXTURE analysis – and we emphasize visualization methods given their importance for large-scale data analysis and interpretation. PCA yields a high-level overview of the patterns of genetic diversity found in a genomic dataset and can be used to delineate genetic ancestry categories [11]. ADMIXTURE can be used to characterize genetic ancestry as a continuous variable, providing fractional estimates of ancestry components for each genomic sample [12].

2. Materials

In order to perform genetic ancestry inference, users will need access to (1) a unix/linux operating system, (2) the Conda package manager and environment management system, (3) program installation files, (4) all necessary program dependencies, and (5) appropriately formatted genomic variant data. We provide an overview of the operating system, the package manger, and the genomic data formats that are needed for both of the genetic ancestry inference methods described here. We also provide details on the installation of the R studio package, which can be used to visualize the results of the genetic ancestry inference.

2.1. Operating Systems

Scientific computing, including genetic ancestry inference, is generally conducted in the command line interface provided by unix/linx operating systems. There are numerous unix/linux operating systems available, many of which are provided free of charge. We recommend the freely available RedHat or Ubuntu Linux operating systems, and all protocols described here can be successfully executed in one of those operating systems.

1. RedHat https://ubuntu.com/download/desktop

2. Ubuntu https://access.redhat.com/downloads

2.2. Package Manager and Environment Management System

Installation and execution of scientific software packages often requires a specific environment along with a number of dependencies, i.e., other programs or libraries. Thus, environment specification and dependency installation is a rate-limiting step for the use of scientific software, including genetic ancestry inference packages. Conda is a freely available software package and environment management system that allows users to install and update software packages and their dependencies. Use of Conda can save a great deal of time and effort, allowing users to focus on software execution without the need for source code compilation. It should be noted that not all genetic ancestry inference software is made available through Conda, users may have to install and compile source files for some packages, but the tools described here can all be installed from Conda.

1. Conda version 4.9.2 https://repo.anaconda.com/miniconda/

2.3. PLINK

The program PLINK v1.90b6.21 64-bit, which can be used for PCA, is distributed through Conda.

1. https://anaconda.org/sjnewhouse/plink

2.4. ADMIXTURE

The program ADMIXTURE version 1.3.0 is distributed through Conda.

1. https://anaconda.org/bioconda/admixture

2.5. Genomic Variant Data

There are numerous sources of genomic variant data, and users can use their own appropriately formatted data to conduct the genetic ancestry inference analyses described here. The 1000 Genomes Project provides freely available human genomic variant data for <2000 individuals from 26 worldwide populations 13. Data are distributed in the variant call format (VCF) and can be downloaded as individual chromosome files. We recommend using a single small chromosome, e.g., chromosome 22, to get started with the analyses described here. Whole-genome analyses can be performed by concatenating all chromosome-specific VCF files.

1. http://1000genomes.ebi.ac.uk/vol1/ftp/data_collections/1000G_2504_high_coverage/working/phase3_liftover_nygc_dir/

   1. phase3.chr22.GRCh38.GT.crossmap.vcf.gz
   2. phase3.chr22.GRCh38.GT.crossmap.vcf.gz.tbi

2.6. Genomic Sample Information

The genetic ancestry inference protocols presented here entail the analysis of a subset of samples from four populations from the 1000 Genomes Project. Users will need to use the metadata on the population origins of the genomic variant samples in order to extract samples from those four populations.

1. https://www.internationalgenome.org/data-portal/sample

2.7. Programming

The R studio version 4.0.3 integrated development environment (IDE) is used to visualize the results of genetic ancestry inference.

1. https://www.rstudio.com/products/rstudio/download/#download

3. Methods

3.1. Software Installation

All of the necessary software listed in the Materials section should be installed, starting with the RedHat or Ubuntu operating system followed by the Conda package manager and environment management system. RedHat is often used for shared computer servers, whereas Ubuntu is recommended for laptop or personal computer use. Conda versions of individual programs can then be installed using the links provided in the Materials section. Finally, the R studio package for results visualization should be installed using the link provided in the Materials section. Instructions for software installation are provided in Notes 1–4.

3.2. Download Genomic Variant Data

Users can use their own genomic variant data as long as it is in the appropriate variant call format (VCF) https://samtools.github.io/hts-specs/VCFv4.1.pdf, or they can download human genome variant data from the 1000 Genomes Project using the link provided in the Materials section. See Note 5.

3.3. Download Genomic Sample Information

Metadata on the population origins of the genomic variants samples from the 1000 Genomes Project can be downloaded using the link in the Materials section. See Note 6.

3.4. Extract Samples from the Four Populations to Be Analyzed

Colombian in Medellin, Colombia [CLM], Iberian Populations in Spain [IBS], Peruvian in Lima Peru [PEL], and Yoruba in Ibadan, Nigeria [YRI]. Please note that the “$” symbol refers to the start of the Linux command line, after which the commands should be entered and executed as shown.

1. Extract sample identifiers for the four populations.

   $ cut -f1,4 igsr_samples.tsv ∣ grep -e “CLM” -e “IBS” -e "PEL" -e "YRI" > samplesToPops.tsv

2. Extract the genomic variant data corresponding to the sample identifiers for the four populations.

   $ cut -f1 samplesToPops.tsv > sampleIDs.tsv

   $ plink --vcf phase3.chr22.GRCh38.GT.crossmap.vcf.gz --keep-allele-order --keep-fam sampleIDs.tsv --make-bed --out 1000Genomes.4Pops. Chr22.GRCh38

The last command creates the three PLINK format files, which are needed to run PCA in PLINK:

1000Genomes.4Pops.Chr22.GRCh38.bed

1000Genomes.4Pops.Chr22.GRCh38.bim

1000Genomes.4Pops.Chr22.GRCh38.fam

3.5. Linkage Disequilibrium (LD) Pruning

Perform linkage disequilibrium (LD) pruning to yield a reduced set of unlinked genetic variants. See Note 7.

1. Filter variants with minor allele frequency of 1% and perform LD pruning.

$ plink --bfile 1000Genomes.4Pops.Chr22.GRCh38 --keep-allele-order --maf0.01 --indep-pairwise 500 5 0.25 --out 1000Genomes.4Pops.Chr22.GRCh38

This command creates the file: 1000Genomes.4Pops.Chr22.GRCh38.prune.in with 20,936 variants.

$ plink --bfile 1000Genomes.4Pops.Chr22.GRCh38 --keep-allele-order --extract 1000Genomes.4Pops.Chr22.GRCh38.prune.in --make-bed --out 1000Genomes.4Pops.Chr22.GRCh38.Pruned

This command creates the files:

1000Genomes.4Pops.Chr22.GRCh38.Pruned.bed

1000Genomes.4Pops.Chr22.GRCh38.Pruned.bim

1000Genomes.4Pops.Chr22.GRCh38.Pruned.fam

3.6. PCA Analysis

Run PCA analysis to characterize the genetic relationships among the samples (see Fig. 1). See Note 8.

Fig. 1.

Principal components analysis (PCA) of four human populations: Colombia (CLM – green), Peru (PEL – red), Spain (IBS – orange), Yoruba (YRI – blue). The first two principal components (PCs) are shown

$ plink --bfile 1000Genomes.4Pops.Chr22.GRCh38.Pruned --pca --out 1000Genomes.4Pops.Chr22.GRCh38.Pruned.PCA

This command creates the PCA results files, which will be subsequently visualized in R studio:

1000Genomes.4Pops.Chr22.GRCh38.Pruned.PCA.eigenval

1000Genomes.4Pops.Chr22.GRCh38.Pruned.PCA.eigenvec

3.7. Visualize PCA Results

Visualize PCA results in R studio.

1. Install R packages needed for visualization.

   install.packages("dplyr")

   install.packages("ggplot2")

   install.packages("reshape2")

2. Configure and import the libraries.

   options(scipen=100, digits=3)

   library(‘dplyr’)

   library(‘ggplot2’)

3. Read and process the eigenvector PCA output file and extract the top two PCs.

   eigenvec <- read.table(‘1000Genomes.4Pops.Chr22.GRCh38.Pruned.PCA.eigenvec’, header = TRUE, sep = ’’)

   rownames(eigenvec) <- eigenvec[,2]

   eigenvec <- eigenvec[,3:ncol(eigenvec)]

   colnames(eigenvec) <- paste(‘PC’, c(1:20), sep = ’’)

   eigenvec$IndividualID = row.names(eigenvec)

   eigenvec = eigenvec[c("IndividualID", "PC1", "PC2")]

   row.names(eigenvec) <- NULL

4. Visually inspect the eigenvector file.

   head(eigenvec)

   	IndividualID	PC1	PC2
   1	HG01113	0.0336	−0.00201
   2	HG01119	0.0384	0.02310
   3	HG01121	0.0371	0.02146
   4	HG01122	0.0277	0.02503
   5	HG01124	0.0283	−0.02970
   6	HG01125	0.0293	−0.03324

5. Merge the results with population group data.

   individualToPopGroupData = read.table(‘samplesToPops.tsv’, col.names = c("IndividualID", "PopGroup"), header = FALSE)

   combinedPCsPopGroups = join(eigenvec, individualToPopGroupData)

   head(combinedPCsPopGroups)

   	IndividualID	PC1	PC2	PopGroup
   1	HG01113	0.0336	−0.00201	CLM
   2	HG01119	0.0384	0.02310	CLM
   3	HG01121	0.0371	0.02146	CLM
   4	HG01122	0.0277	0.02503	CLM
   5	HG01124	0.0283	−0.02970	CLM
   6	HG01125	0.0293	−0.03324	CLM

6. Define population colors.

   colors <- c("CLM" = "green", "IBS" = "orange", "PEL" = "red", "YRI" = "blue",)

7. Plot the PCA results for the top two PCs.

   Options(repr.plot.width= 12, repr.plot.height=12)

   ggplot(combinedPCsPopGroups, aes(x=PC1, y=PC2, fill=PopGroup)) + scale_fill_manual(values=colors) +

   geom_point(size = 6, pch = 21, color= "black") + theme_classic() + theme(axis.title=element_text(size=18))

3.8. Admixture

Run ADMIXTURE analysis to characterize patterns of genetic ancestry and admixture among the samples (see Fig. 2). See Note 9.

Fig. 2.

ADMIXTURE plot of four human populations: Colombia (CLM), Spain (IBS), Peru (PEL), and Yoruba (YRI). Each column is an individual, and for each individual the ancestry fraction for each of three continental population groups is shown: African (blue), European (orange), and Native American (red)

$ admixture 1000Genomes.4Pops.Chr22.GRCh38.Pruned.bed 3-j4

This command creates the files:

1000Genomes.4Pops.Chr22.GRCh38.Pruned.3.Q

1000Genomes.4Pops.Chr22.GRCh38.Pruned.3.P

3.9. Visualize ADMIXTURE Results

Visualize ADMIXTURE results using R studio.

1. Install R packages needed for visualization.

   install.packages("dplyr")

   install.packages("ggplot2")

   install.packages("reshape2")

2. Import libraries.

   library("dplyr")

   library("ggplot2")

   library("reshape2")

3. Get individual identifiers and individual population labels for all samples.

   individualToPopGroupData = read.table(‘samplesToPops.tsv’, col.names = c("IndividualID", "PopGroup"), header = FALSE)

   allCombinedFAMData = read.table(‘1000Genomes.4Pops.Chr22.GRCh38.Pruned.fam’, header = FALSE, col. names = c("FamilyID", "IndividualID", "N1", "N2", "N3", "N4"))

   allCombinedFAMData = data.frame("IndividualID" = all-CombinedFAMData[ , 2:2])

4. Visually inspect the results.

   head(individualToPopGroupData)

   	IndividualID	PopGroup
   1	HG01250	CLM
   2	HG01255	CLM
   3	HG01274	CLM
   4	HG01279	CLM
   5	HG01123	CLM
   6	HG01130	CLM

   head(allCombinedFAMData)

   IndividualID

   1	HG01112
   2	HG01113
   3	HG01119
   4	HG01121
   5	HG01122
   6	HG01124

5. Assign population labels to samples from ADMIXTURE output.

   allCombinedFAMandPopGroupData = join(allCombined-FAMData, individualToPopGroupData)

   head(allCombinedFAMandPopGroupData)

   	IndividualID	PopGroup
   1	HG01112	CLM
   2	HG01113	CLM
   3	HG01119	CLM
   4	HG01121	CLM
   5	HG01122	CLM
   6	HG01124	CLM

6. Extract ancestry estimates obtained from ADMIXTURE.

   admixtureAncestryEstimates = read.table("1000Genome-s.4Pops.Chr22.GRCh38.Pruned.3.Q", header = FALSE)

   head(admixtureAncestryEstimates)

   	V1	V2	V3
   1	0.00001	0.00001	1.000
   2	0.00001	0.33841	0.662
   3	0.00001	0.49758	0.502
   4	0.00001	0.49336	0.507
   5	0.06223	0.48273	0.455
   6	0.02341	0.16010	0.816

7. Combine individual ancestry estimates with individual identifiers and population labels.

   combinedAncestryEstimatesData = cbind(allCombinedFA-MandPopGroupData, admixtureAncestryEstimates)

   head(combinedAncestryEstimatesData)

   	IndividualID	PopGroup	V1	V2	V3
   1	HG01112	CLM	0.00001	0.00001	1.000
   2	HG01113	CLM	0.00001	0.33841	0.662
   3	HG01119	CLM	0.00001	0.49758	0.502
   4	HG01121	CLM	0.00001	0.49336	0.507
   5	HG01122	CLM	0.06223	0.48273	0.455
   6	HG01124	CLM	0.02341	0.16010	0.816

8. Characterize population ancestry means.

   options(scipen = 10000)

   combinedAncestryEstimatesData %>% group_by(PopGroup) %>% summarise_at(vars(V1, V2, V3), funs(mean))

   PopGroup	V1	V2	V3
   CLM	0.0709	0.273122	0.65601
   IBS	0.0028	0.001514	0.99569
   PEL	0.0343	0.819700	0.14604
   YRI	0.9991	0.000161	0.00078

9. Rename the columns and assign proper population labels to each cluster.

   combinedAncestryEstimatesData = combinedAncestryEstimatesData %>% dplyr::rename("European" = "V3", "African" = "V1", "NativeAmerican" = "V2",)

   head(combinedAncestryEstimatesData)

   	IndividualID	PopGroup	African	NativeAmerican	European
   1	HG01112	CLM	0.00001	0.00001	1.000
   2	HG01113	CLM	0.00001	0.33841	0.662
   3	HG01119	CLM	0.00001	0.49758	0.502
   4	HG01121	CLM	0.00001	0.49336	0.507

10. Check average ancestry for each population.

    options(scipen = 10000)

    combinedAncestryEstimatesData %>% group_by(PopGroup) %>% summarise_at(vars(European, African, NativeAmerican), funs(mean))

    PopGroup	European	African	NativeAmerican
    <fct>	<dbl>	<dbl>	<dbl>
    CLM	0.65601	0.0709	0.273122
    IBS	0.99569	0.0028	0.001514
    PEL	0.14604	0.0343	0.819700
    YRI	0.00078	0.9991	0.000161

11. Reformat the data in one column for graphs.

    combinedAncestryEstimatesDataSorted = arrange(combinedAncestryEstimatesData, European, African, NativeAmerican, group_by = PopGroup)

    row.names(combinedAncestryEstimatesDataSorted) <- NULL

    combinedAncestryEstimatesDataSorted$index= as.numeric(row-names(combinedAncestryEstimatesDataSorted))

    combinedAncestryEstimatesDataSortedMelt = melt(data = combinedAncestryEstimatesDataSorted, id.vars = c("IndividualID", "PopGroup", "European", "African", "NativeAmerican", "index"), measure.vars = c("European", "African", "NativeAmerican"))

    colnames(combinedAncestryEstimatesDataSortedMelt)[7] <- ‘Ancestry’

    colnames(combinedAncestryEstimatesDataSortedMelt)[8] <- ‘AncestryFraction’

    print(combinedAncestryEstimatesDataSortedMelt[c(‘IndividualID’, "PopGroup", "Ancestry", "AncestryFraction", "index")])

12. Define the ancestry colors.

    colors <- c("African" = "blue", "European" = "orange", "NativeAmerican" = "red")

13. Render the ADMIXTURE plot.

    options(repr.plot.width=16, repr.plot.height=8)

    ggplot(data=combinedAncestryEstimatesDataSortedMelt, aes(x=as.character(index), y=AncestryFraction, fill=Ancestry)) +

    geom_bar(stat= "identity", width=1) + facet_grid(cols = vars(PopGroup), scales = "free", space = "free", drop = TRUE, switch= "both") +

    scale_fill_manual(values=colors) +

    labs(y= "Percentage Ancestry Estimate", title = "Admixture plot for four population groups from 1KGP", subtitle = "Groups: IBS, PEL, YRI, & CLM") +

    theme(axis.title.x=element_blank(), axis.text.x=element_blank(), axis.ticks.x=element_blank(), panel.spacing.x=unit(1, "lines"),

    strip.background = element_blank(), panel.background = element_blank(), axis.text=element_text(size=12),

    axis.title=element_text(size=18), strip.text.x = element_text(size = 18), title = element_text(size = 20))

4. Notes

1. Operating system installation

   We recommend using the Ubuntu operating system if users are working on their own laptop. Instructions for installing Ubuntu on a laptop can be found at https://ubuntu.com/tutorials/install-ubuntu-desktop#1-overview.

2. Conda installation

   Conda can be installed using the Linux commands shown below, following the series of prompts after the first command:

   $ bash Miniconda3-py38_4.9.2-Linux-x86_64.sh

   $ conda create --name myenv

   $ conda activate myenv

   Please note that the “$” symbol refers to the start of the Linux command line, after which the commands should be entered and executed as shown.

3. PLINK and ADMIXTURE installation

   The Conda versions of the programs PLINK and ADMIXTURE can be installed using the Linux commands shown below.

   $ conda install -c sjnewhouse plink

   $ conda install -c bioconda admixture

4. R studio

   R studio can be installed following the instructions in the link provided in the Materials section, using the *.deb file corresponding to the Ubuntu operating system.

5. 1000 Genomes Project human variant data

   The 1000 Genomes Project human genome variant data can be downloaded using the following commands:

   $ wget ftp://ftp.1000genomes.ebi.ac.uk/vol1/ftp/data_collections/1000G_2504_high_coverage/working/phase3_liftover_nygc_dir/phase3.chr22.GRCh38.GT.crossmap.vcf.gz.tbi

   $ wget ftp://ftp.1000genomes.ebi.ac.uk/vol1/ftp/data_collections/1000G_2504_high_coverage/working/phase3_liftover_nygc_dir/phase3.chr22.GRCh38.GT.crossmap.vcf.gz

   Please note that the wget command and the following ftp address should be entered on a single line.

6. The 1000 Genomes Project human genome sample

   The 1000 Genomes Project human genome sample metadata can be downloaded using the “Download the list” button on the link provided in the Materials section. It should be noted that the sample metadata file igsr_samples.tsv will be downloaded into the user’s browser default download directory and will need to be transferred to the Linux working directory where the analyses are to be conducted.

7. Linkage disequilibrium (LD) pruning

   Linkage disequilibrium (LD) pruning yields a reduced subset of genetic variants that are in approximate linkage equilibrium with each other (i.e., unlinked variants). The parameter values for the --indep-pairwise flag define how LD pruning proceeds. The first value (500) refers to the size of the genomic window, the second value (5) is the window step size, and the third value (0.25) is the variance inflation factor, which is used to measure the extent of linkage between pairs of variants. Details on LD pruning in PLINK can be found at https://zzz.bwh.harvard.edu/plink/summary.shtml#prune.

8. PCA analysis

   Principal components analysis (PCA) is a technique that is used to reduce the dimensionality of large datasets and is ideal for the analysis of high-dimension genomic variant data. PCA analyzes the genomic variant covariance matrix to create uncorrelated, orthogonal variables that maximize the variance in the data – the principal components (PCs). Visualizing genomic data in PC-space provides an intuitive way to evaluate the genetic relationships among samples. The PLINK PCA program yields 20 PCs, and we visualize the first two PCs here.

9. ADMIXTURE

   The program ADMIXTURE characterizes each individual with respect to the proportion of the genome that is made up of K theoretical ancestral source populations. The value of K can be chosen a priori, based on knowledge of the populations under analysis, or ADMIXTURE can be run across a series of K values to determine which value fits the data best. The samples analyzed here are from Africa (YRI), Europe (IBS), and the Americas (CLM and PEL). Since these populations are characterized primarily by African, European, and Native American continental ancestry fractions, we chose a value of K = 3 for the analysis. The value of 3 after the file name in the command line corresponds to the K-value, and the −j 4 flag corresponds to the number of threads to use.