Analysis code accompanying the manuscript "Improving PGS for Lpa(a) and CAD with LPA-augmented LD panels"

---

Overview

This repository contains R code used for the study "Improving Polygenic Risk Scores for Lipoprotein(a) and Coronary Artery Disease with LPA-augmented LD panels".

This study is not yet published (under development/review).

Description of files in the data/ folder:

• data/snplist.snplist:
  curated set of 14,113 variants in the LPA gene region (6:159452514-162585307 on hg19/GRCh37)

• data/composite_variants_hm3_plus_LPA_LD_panel_HRC.txt:
  composite list of variants of the HM3+ LD panel published by Privé et al., and the custom high-density LD panel (cf. variant list above), while restricting to variants present in the HRC r1.1 2016 reference panel

Both variant lists are in format "CHR:POS:REF:ALT" without header and with POS on genome build hg19/GRCh37.

---

Requirements

1. Clone the repository

   git clone https://github.com/genepi/LPA_LD_panel_public.git
   cd LPA_LD_panel_public

2. Install dependencies

   conda env create -f bigsnpr_env.yml
   conda activate bigsnpr

3. Download GWAS summary statistics and LD reference panels

   GWAS summary statistics:

   • CAD GWAS published by Aragam et al.
   • Lp(a) GWAS (will be available upon publication of this study)

   LD panels:

   • HM3 published by Privé et al.
   • HM3+ published by Privé et al.

NOTE: For some analyses (e.g. de-novo calculation of the custom high-density LD panel for the LPA gene region) individual-level data from the UKB is required.

---

Code Execution

Run scripts from terminal with Rscript in activated conda environment (recommended).

• Compute custom high-density LD reference panel for the LPA gene region:

   Rscript src/01_panel_dev/01_generate_custom_LPA_LD_panel.R  --bgen input.bgen --snplist data/snplist.snplist [--outdir OUTPUT] [--cores nr_cores]

• Combine custom LPA LD panel with chr6 of public HM3+ LD panel:

   Rscript src/01_panel_dev/02_merge_panels_chr6.R --corr0 x.LD_corr0.rds --ldscores x.LD_scores.rds --corr0_ref_chr6 LD_with_blocks_chr6.rds --ref_meta map_hm3_plus.rds [--snplist data/snplist.snplist] [--outdir OUTPUT]

• Create de-novo PGS with LDpred2-auto (one example):

   Rscript src/02_score_dev/LPA/01_construct_LPA_score_hm3.R --sumstats LPA_sumstats.regenie.gz --ldref_folder data/LD_ref/hm3/ [--outdir OUTPUT] [--cores nr_cores]

• Compute PGS values in a target cohort:

   Rscript src/03_score_calc/bigsnpr_calc_score_from_bgen.R --score score.txt --bgenfiles bgen_paths.txt [--bgensample geno.sample] [--targetsamples target_sample_ids.txt] [--outdir OUTPUT] [--cores nr_cores]

• Get model, which can be used to calibrate Lp(a) PGS to the original measurement scale for Lp(a):

   Rscript src/04_score_calib/LPA_scores_calibration_UKB_test_set.R --pheno UKB.txt --samples IDs.txt --score xxx.scores.txt [--outdir OUTPUT]

For all scripts a message describing the input requirements can be generated like this:

   Rscript src/01_panel_dev/01_generate_custom_LPA_LD_panel.R --help

---

Contact

This code was developed at the Institute of Genetic Epidemiology, Medical University of Innsbruck

• Florian Schnitzer
• Lukas Forer (@lukfor)