gwas_ldsc_tutorial.txt

Instructions for Day 4. LD-score regression.

- Download summary statistics for Schizophrenia GWAS from Psychiatric Genomics Consortium 
  1. Open browser and go to https://www.med.unc.edu/pgc/results-and-downloads
  2. Search for SCZ2 in the first column. Select "Download full SNP results" link from the second column.
  3. Read and accept Data Download Agreement. Click Submit. On the next page, click "Download SCZ full SNP data".
  4. You should receive a file called ckqny.scz2snpres.gz . Save it in your favorite location (for exampel,  ~/gwas_course )

- Download and install Anaconda Python distribution 
  1. Open browser and go to https://www.anaconda.com/distribution/
  2. Select your operating system (mac, windows or linux)
  3. Click "Download" button under "Python 3.7 installer"
  4. Install the downloaded package. This may require administrator rights on your machine.

- Open terminal, and execute "python" command. Then, execute the following:

# Load modules
import pandas as pd
import numpy as np
from scipy import stats

# Read summary statistics table from PGC schizophrenia GWAS.
# If 'ckqny.scz2snpres.gz' is not in your current folder you may need to provid full path.
df=pd.read_table('ckqny.scz2snpres.gz', delim_whitespace=True)

# Display header of the table
df.head()

# Rename columns as follows: hg19chrc -> CHR, snpid -> SNP, etc.
df.rename(columns={'hg19chrc':'CHR', 'snpid':'SNP', 'a1':'A1', 'a2':'A2', 'or':'OR', 'se':'SE', 'p':'PVALUE', 'bp':'BP', 'info':'INFO'}, inplace=True)

# Drop 'ngt' column from the table
df.drop(columns=['ngt'], inplace=True)

# Create new columns 'N' and 'Z' according to specified formulas
df['N'] = 35476+46839
df['Z'] = -stats.norm.ppf(df['PVALUE'].values*0.5)*np.sign(np.log(df['OR'].values))

# Simplify format in the 'CHR' column, and convert it from string to numeric representation
df['CHR'] = pd.to_numeric(df['CHR'].str.replace('chr', ''), errors='coerce')

# Drop rows of the table that correspond to non-numeric chromosomes
df = df[~df['CHR'].isnull()].copy()
df['CHR']=df['CHR'].astype(int)

# Save the resulting table into file 'PGC_SCZ_2014'
df[['CHR', 'SNP', 'A1', 'A2', 'BP', 'INFO', 'PVALUE', 'N', 'Z']].to_csv('PGC_SCZ_2014', index=False, sep='\t')

- Open another terminal and compress the resulting file:

zip PGC_SCZ_2014.zip PGC_SCZ_2014

- Optional exercise:

  Download 'http://ldsc.broadinstitute.org/static/media/w_hm3.noMHC.snplist.zip' and put it in the same folder as 'ckqny.scz2snpres.gz'

  Start 'python' and execute the following:

ref=pd.read_table('w_hm3.noMHC.snplist.zip', delim_whitespace=True)
df_ref = pd.merge(df, ref[['SNP']], on='SNP', how='inner')
df_ref[['CHR', 'SNP', 'A1', 'A2', 'BP', 'INFO', 'PVALUE', 'N', 'Z']].to_csv('PGC_SCZ_2014_hm3', index=False, sep='\t')

- Open another terminal and compress the resulting file:

zip PGC_SCZ_2014_hm3.zip PGC_SCZ_2014_hm3

- Now we can run LD score regression. 
  Open browser and go to http://ldsc.broadinstitute.org/

  Click "Get started with LD hub"
  Click "Sign in with your google account"
  Click "Go to test center"
  Click "Continue"
  Click "Choose file", and and select "PGC_SCZ_2014_hm3.zip" file for upload
  Click "Submit"
  Download QC information for the uploaded file
  Download SNP heritability analysis results