Guideline to use RFmix for local ancestry inference

Warning

This guideline assume you have phased genotype in PLINK2 pgen format of your own dataset. PLINK2 can convert many other formats to pgen format (see here). If you need help, raise an issue so we can help update this guideline to adapt to your data.

Note

Note on converting UK Biobank version 2 phased haplotype data (contributed by Jordan Rossen from HSPH):

When converting UK Biobank version 2 phased haplotype data bgen format using plink2, for chromosomes 1-20, the ref-first flag should be used (--bgen ref-first) and for chromosomes 21-22, the ref-last flag should be used (--bgen ref-last). This is because the allele encoding is switched for chromosomes 21 and 22, specifically in the phased haplotype data (this is not an issue in the UK Biobank imputed data). This issue has been previously noted on the UK Biobank mailing list (registration is needed).

Step 1: download and process reference data

admix get-1kg-ref --dir=/path/to/1kg --build hg38
# or --build hg19

Note

This script will (1) download plink2 pgen files for 1,000 genomes project (2) perform basic QCs (3) download metadata of sample map and genetic maps. This will take a while to process (~2hrs). You can also read more details behind the scene.

As RFmix takes reference dataset in vcf format, we convert pgen format to vcf format

# convert plink2 to vcf
root_dir=/path/to/1kg/
mkdir -p ${root_dir}/vcf/
for chrom in {1..22}; do
    plink2 \
    --pfile ${root_dir}/pgen/all_chr \
    --export vcf bgz \
    --chr ${chrom} \
    --out ${root_dir}/vcf/chr${chrom} && tabix -p vcf ${root_dir}/vcf/chr${chrom}.vcf.gz
done

Step 1.5 (optional) subset admixed individuals using joint PCA with 1KG reference

We first merge the 1kg data and sample data into a single file and perform a joint PCA.

ref_pfile=/path/to/1kg/all_chr    # path to 1kg pgen file (all chromosomes)
sample_pfile=/path/to/sample/all_chr # path to sample pgen file (all chromosomes)
out_dir=/path/to/joint-pca    # path to output directory

# merge 1kg dataset and sample dataset
admix pfile-merge-indiv \
    --pfile1 ${ref_pfile} \
    --pfile2 ${sample_pfile} \
    --out ${out_dir}/merged

# perform PCA (you may want to remove the 'approx' modifier if you don't have many samples in your data set (<5,000 individuals))
plink2 --bfile ${out_dir}/merged \
    --pca approx \
    --out ${out_dir}/merged_pca

Then we plot the joint PCA results.

admix plot-joint-pca \
    --ref-pfile ${ref_pfile} \
    --pca-prefix ${out_dir}/merged_pca \
    --out ${out_dir}/merged_pca

And we select admixed individuals based on PC.

admix select-admix-indiv \
    --ref-pfile ${ref_pfile} \
    --pca-prefix ${out_dir}/merged_pca \
    --superpop1 EUR --superpop2 AFR \
    --exclude-pop2 ASW,ACB \
    --out ${out_dir}/selected_admix

And create new plink2 files

for chrom in $(seq 1 22); do
    plink2 \
        --pfile /path/to/sample/chr${chrom} \
        --keep ${out_dir}/selected_admix.indiv \
        --make-pgen --out /path/to/sample/admix/chr${chrom} \
        --maf 0.005 \
        --memory 24000
done

Step 2: run RFmix

Step 2.1: download and compile RFmix

Follow https://github.com/slowkoni/rfmix#building-rfmix to build the rfmix executables.

Step 2.2: run RFmix

Now we will run RFmix to infer local ancestry for your genotype file (see how to prepare the genotype file in Prepare dataset section).

## Specify constants
chrom=XX # this script should be ran 1 chromosome at a time
REF_DIR=/path/to/1kg/ # see step 1, set it to the root directory containing metadata/ pgen/ vcf/
RFMIX=/path/to/rfmix_exe # see above to download and compile rfmix executables

pfile=/path/to/your-plink2-file # see "Prepare dataset" section
out_prefix=/path/to/output/chr${chrom} # output prefix

## convert pfile to vcf because RFmix takes vcf as input
plink2 \
    --pfile ${pfile} \
    --chr ${chrom} \
    --output-chr 26 \
    --export vcf bgz \
    --out ${out_prefix}.tmp

tabix -p vcf ${out_prefix}.tmp.vcf.gz

## Filter reference samples
# extract 1st column (sample name) and 2nd column (populations)
# Here, we use CEU, YRI, PEL for reference populations for Latino populations
# NOTE: it is up to your choice what reference populations to use
# e.g., For local ancestry inference of European-African admixed individuals, we 
# recommend using CEU, YRI only.
awk '$2=="CEU" || $2=="YRI" || $2=="PEL" {print $1 "\t" $2}' \
    ${REF_DIR}/metadata/unrelated_sample.tsv >${out_prefix}.tmp.sample_map.tsv

## run RFmix
${RFMIX} \
    -f ${out_prefix}.tmp.vcf.gz \
    -r ${REF_DIR}/vcf/chr${chrom}.vcf.gz \
    -m ${out_prefix}.tmp.sample_map.tsv \
    -g ${REF_DIR}/metadata/genetic_map/chr${chrom}.tsv \
    --chromosome=${chrom} \
    -o ${out_prefix}

## Remove temporary files
# .sis file corresponds to XX
# .fb file corresponds to XX
rm ${out_prefix}*tmp*
rm ${out_prefix}.sis.tsv
rm ${out_prefix}.fb.tsv

After you finish this step, you can proceed to the rest of preparing the dataset.