Practical 5: Using LDSC to estimate heritability and genetic correlations

This runs on the GGWS server.

We provide summary statistics for two GWAS of Waist-to-Hip Ratio (WHR) in males and females conducted by the GIANT consortium.

Preparation

Go to /scratch/module3/ and create a directory where to run the practical and copy GWAS summary statistics there

cd
cd module3
mkdir prac5/
cd prac5
cp /data/module3/Prac5_LDSC/GIANT_2015_WHR_*_EUR.txt .

Load LDSC software and required libraries (more details available on github: https://github.com/bulik/ldsc)

conda activate ldsc

Format GWAS summary statistics for LDSC analysis

 /software/ldsc/munge_sumstats.py \
  --sumstats GIANT_2015_WHR_FEMALES_EUR.txt \
  --merge-alleles /data/module3/Prac5_LDSC/eur_w_ld_chr/w_hm3.snplist --chunksize 1000000 \
  --out giant_whr_females


 /software/ldsc/munge_sumstats.py \
  --sumstats GIANT_2015_WHR_MALES_EUR.txt \
  --merge-alleles /data/module3/Prac5_LDSC/eur_w_ld_chr/w_hm3.snplist --chunksize 1000000 \
  --out giant_whr_males

Part 1: Estimation of heritability using LDSC

Run the following commands

/software/ldsc/ldsc.py --h2 giant_whr_females.sumstats.gz \
  --ref-ld-chr /data/module3/Prac5_LDSC/eur_w_ld_chr/ \
  --w-ld-chr /data/module3/Prac5_LDSC/eur_w_ld_chr/ \
  --out h2_giant_whr_females

/software/ldsc/ldsc.py --h2 giant_whr_males.sumstats.gz \
  --ref-ld-chr /data/module3/Prac5_LDSC/eur_w_ld_chr/ \
  --w-ld-chr /data/module3/Prac5_LDSC/eur_w_ld_chr/ \
  --out h2_giant_whr_males

Question 1

How many SNPs were used in each regression?
What is the heritability of WHR in males and females?
What is the LD score intercept? Is this expected? If not then what could explain this observation?

Part 2: Estimation of genetic correlation using LDSC

Run the following command

/software/ldsc/ldsc.py --rg giant_whr_females.sumstats.gz,giant_whr_males.sumstats.gz \
  --ref-ld-chr /data/module3/Prac5_LDSC/eur_w_ld_chr/ \
  --w-ld-chr /data/module3/Prac5_LDSC/eur_w_ld_chr/ --out rg_giant_whr

Question 2

How many SNPs were used in each regression?
What is the heritability of WHR in males and females? Is this different from your previous results? Rerun the LDSC commands from Question 1 adding the following flag: --two-step INFINITY. What can you conclude?
Interpret the bivariate LD score intercept.
What can you conclude regarding the genetic architecture of WHR in males and females.

Part 3: Estimation of genetic correlation using LDSC

We now provide summary statistics for two GWAS of Waist-to-Hip Ratio (WHR) in males and females in the UK Biobank (UKB). The files are already formatted for LDSC.

Copy the new GWAS summary statistics

cp /data/module3/Prac5_LDSC/ukb_whr_*.sumstats.gz .

Question 3

Estimate the heritability of WHR in males and females from the UKB. How does it compare with your previous results. Is this expected?
Estimate the genetic correlation between UKB males and UKB females for WHR. How does it compare with you previous analyses using data from GIANT?
Interpret the bivariate LDSC intercept for the two WHR GWAS in the UK Biobank.
Estimate the genetic correlation between UKB (fe)males and GIANT (fe)males for WHR.

Appendix: How to download LDSC (at home)

Download LDSC files

git clone https://github.com/bulik/ldsc

Download LDSC scores calculated in European ancestries individuals from the 1000 Genomes Project

wget https://data.broadinstitute.org/alkesgroup/LDSCORE/eur_w_ld_chr.tar.bz2
tar -xvf eur_w_ld_chr.tar.bz2