Recipes for GWAS data conversion/extraction

Useful tools

• Download PLINK 1.9 from https://www.cog-genomics.org/plink2.

• Install bcftools from https://samtools.github.io/bcftools/ but for MAC OS X it is easier to install UNIX tools via homebrew. After having homebrew installed, just tap in by brew install bcftools.

• Download SNP annotation for hg19 (see below)

wget ftp://ftp.ncbi.nih.gov/snp/organisms/human_9606_b151_GRCh37p13/VCF/common_all_20180423.vcf.gz

Example: Downloading UK biobank genotype data using virtual box

Since UK Biobank genotypes are only downloadable by a linux-compiled binary, I had to spin off a virtual box. I thought it is generally a good idea to have a shared directory between the host (Mac OS) and the box. Here is how I did based on the stackoverflow search.

0. Create a virtual box with ICH9 chipset and the bridged network adapter.

1. Install an extension pack (Oracle_VM_Virtual Box etc.).

2. Run “Insert Guest Additions CD images…” under the “Devices” tab.

3. Restart the box.

4. Create a shared folder via GUI

5. Mount the folder

mount -t vboxsf mountfolder /home/devInHost/mountfolder

Once I have successfully mounted the shared folder, I can use ukbgene to download the imputed genotypes. To make sure that I write down to the shared folder, I just run the command with the root priviledge.

parallel -j4 ./ukbgene imp -c{1} :: {1..22}

Enjoy!

Example: Extracting dosage information from a VCF file to use a PrediXcan model

• Download weights Brain_Amygdala.tar.gz from http://predictdb.org/.

• Copy *.py and *.R from PrediXcan github repository.

1. We need to construct annotations

cat common_all_20180423.vcf.gz | gzip -d | bgzip -c > snps.vcf.gz && tabix snps.vcf.gz

2. Annotate rsID to match with the DB file.

We will store the updated VCF file in separate directory:

mkdir -p data/vcf/

We can create annotated VCF.

bcftools annotate -c CHROM,FROM,TO,ID -a snps.vcf.gz -o data/vcf/chr21.vcf.gz chr21.dose.vcf.gz

3. Create dosage file as PrediXcan requires

mkdir -p data/dosage/

We can extract dosage information with the format -f "[\t%DS]\n", prepending 6 header columns: chromosome, rsID, position, reference, alternative allele, and minor allele frequency with the format -f "%CHROM\t%ID\t%POS\t%REF\t%ALT\t%MAF". To make sure that our prediction is reliable, we may filter out SNP with MAF less than 5% adding -e "MAF[0]<0.05".

$ bcftools query -e "MAF[0]<0.05" -f "%CHROM\t%ID\t%POS\t%REF\t%ALT\t%MAF[\t%DS]\n" data/vcf/chr21.vcf.gz  | gzip > data/dosage/chr21.dosage.gz

Additionally we can list samples in the same directory:

$ bcftools query -l data/vcf/chr21.vcf.gz | awk -F'\t' '{ print $1 FS $1 }' > data/dosage/chr21.samples

Once I have successfully mounted the shared folder, I can use ukbgene to download the imputed genotypes. To make sure that I write down to the shared folder, I just run the command with the root priviledge.

parallel -j4 ./ukbgene imp -c{1} :: {1..22}

4. Run PrediXcan.

python PrediXcan.py --predict --dosages data/dosage/ --dosages_prefix chr21 --output_prefix temp --samples chr21.samples --weights Brain_Amygdala/gtex_v7_Brain_Amygdala_imputed_europeans_tw_0.5_signif.db

Conclusion

• bcftools are just amazing.
• I found the way PrediXcan process files quite inefficient.