OAPRS

ADNI adjusted IGAP summary

ADNI adjusted IGAP summary

You can download summary statistics generated using OAPRS on The International Genomics of Alzheimer's Project (IGAP)
with sample overlap adjustment of Alzheimer's Disease Neuroimaging Initiative (ADNI) genotypes.

Description

OAPRS is designed to help and guide adjusting sample overlap bias in building PRS without overfitting.
OAPRS consists of four main steps: 1.summary information preparation, 2.sample overlap adjustment, 3.PRS construction, and 4.validation using visual diagnostics.
Before the preparation step, GWAS summary statistics using only overlapped individual genotypes from the target data need to be generated using standard GWAS softwares.
OAPRS assumes target data to be a PLINK binary file format.

Detailed Manual

For list of functions and detailed options in OAPRS, Please refer to this manual : OAPRS Manual

Dependencies

data.table, dplyr, ggplot2, RcppArmadillo, Rcpp (>= 1.0.9)

R.utils are necessary for reading gzipped files install.packages('R.utils')

Installation

You can install OAPRS Package from OAPRS github using r-devtools

devtools::install_github('leelabsg/OAPRS')

Example Usage

Example files are found in extdata of OAPRS package repository.

1. Format summary statistics

Let's read a partial sample summary statistics on large scale genetic consortium with Check_Sums.

data_path = system.file("extdata/example",package = "OAPRS")

Create Column names first. By setting the cols variable, OAPRS reformat summary statistics for further use.

cols = c(BETA="beta",Pval="pval",CHR="chrom",POS="pos",REF="ref",ALT="alt",SNP="rsids")

We can designate the summary statistics file path and genome_build, and population. If there is not specified column for sample size, Spcf_n, Spcf_n_case, Spcf_n_ctrl are needed.

cs = Check_Sums(paste0(data_path,'/consortium.ss'),
Genome_Build = "hg37", Pop = "eas",
cols=cols,
Spcf_n=249625,Spcf_n_case = 50466, Spcf_n_ctrl = 199159)

Similarly, we can format summary statistics with target summary.

cols = c(BETA="beta",Pval="pval",CHR="chrom",POS="pos",REF="ref",ALT="alt",SNP="rsids",SE="sebeta")

ts = Check_Sums(paste0(data_path,'/target.ss'),
Genome_Build = "hg37", Pop = "eas",
cols=cols,
Spcf_n=72210,Spcf_n_case = 5083, Spcf_n_ctrl = 62127)

2. Adjust consortium and target gwas

With formatted summary statistics cs and ts, we can build adjusted summary statistics using exclude_overlap.

adj_ss = exclude_overlap(cs,ts,"adj.txt",phenotype="binary")

3. (External) Build PRS from prscs, lassosum, ldpred2, etc.

In this example, prscs is applied In order to run prscs, we subset adjusted summary as input format.

library(dplyr)
write.table(adj_ss %>% select(SNP,A1=ALT,A2=REF,BETA=BETA_all,P=P_all),paste0(data_path,"/adj_ss_all.txt"),quote = F, col.names = T, row.names = F)
write.table(adj_ss %>% select(SNP,A1=ALT,A2=REF,BETA=BETA_IVW,P=P_IVW),paste0(data_path,"/adj_ss_IVW.txt"),quote = F, col.names = T, row.names = F)
write.table(adj_ss %>% select(SNP,A1=ALT,A2=REF,BETA=BETA_RZ,P=P_RZ),paste0(data_path,"/adj_ss_RZ.txt"),quote = F, col.names = T, row.names = F)

Here is some example usage for prscs for each adjustment methods.

for i in all IVW RZ
do
PRScs.py \
--ref_dir=snpinfo_1kg_hm3_eas.gz \
--bim_prefix=target \
--sst_file=adj_ss_${i}.txt \
--n_gwas=177415 \
--out_dir=${outdir}/${i}
done

4. Make filters for visual diagnostics

Genome build and population information is essential in variant filtering for visual diagnostics.

lds = Marker_select_ld(adj_ss, Genome_Build="hg37", Pop = "eas")

5. Evaluate Scores on each thresholds

With result from 4 and prs weights on 3, scores will be evaluated by corresponding sample ID's. You can specify platform for prs weights or specify column names

prs_res_paths=c(paste0(data_path,'/prscs_all_chr_unadj.txt'),
  paste0(data_path,'/prscs_all_chr_IVW.txt'),paste0(data_path,'/prscs_all_chr_RZ.txt'))
scr = score_eval(prs_res_paths,lds, target_path = paste0(data_path,"/target"),platform="prscs",
  pheno_path = paste0(data_path,"/target.pheno"),ID_col="V2",pheno_col="V3")

For utilizing individual prs, columns with "ref" in the result can be used.

scr %>% select(IID, contains('ref'))

6. Generate Visual Diagnostics

Lets make visual diagnostics plot with generated scores. the legends are set using method_names option

diagnostic_plt(score=scr,title="test",
  Output_Plot_path="~/plot.png",method_names = c("all","IVW","RZ"))