PRSfromPGS

Calculate Polygenic Risk Scores (PRS) from genetic data using PGS Catalog-format scoring models and bgzipped VCF files.

Installation

PRSfromPGS depends on Bioconductor packages. Install them first, then install PRSfromPGS from GitHub:

# Install Bioconductor dependencies
if (!requireNamespace("BiocManager", quietly = TRUE))
    install.packages("BiocManager")
BiocManager::install(c("VariantAnnotation", "Rsamtools"))

# Install PRSfromPGS
if (!requireNamespace("remotes", quietly = TRUE))
    install.packages("remotes")
remotes::install_github("USCbiostats/PRSfromPGS")

For parallel processing support, also install:

# Parallel chromosome processing
install.packages(c("future", "future.apply", "parallelly"))

# Parallel VCF indexing
BiocManager::install("BiocParallel")

Quick Start

library(PRSfromPGS)

# Step 1: Prepare VCF files (creates per-chromosome RDS summaries)
vcf_files <- list.files("data/", pattern = "\\.vcf\\.gz$", full.names = TRUE)
prepareVCFfiles(vcf_files, genome = "hg38", tabix = TRUE)

# Step 2: Read a PGS Catalog scoring file
pgs_model <- readPGSmodel("PGS000001.txt")

# Step 3: Calculate PRS across all autosomes
result <- calculatePRS(pgs_model)
prs_values <- result$prs  # Named numeric vector (one score per sample)

Usage

Preparing VCF Data

VCF files must be bgzipped (.vcf.gz). Each file should contain variants for a single chromosome.

# Index and summarize VCF files
vcf_files <- c("chr1.vcf.gz", "chr2.vcf.gz", "chr3.vcf.gz")
prepareVCFfiles(vcf_files, genome = "hg38", tabix = TRUE)

# This creates chromosome1VCFsummary.rds, chromosome2VCFsummary.rds, etc.
# in the current working directory

Reading PGS Models

Reads tab-delimited scoring files from the PGS Catalog. Handles harmonized columns (hm_chr, hm_pos, hm_inferOtherAllele) automatically.

pgs_model <- readPGSmodel("PGS000001.txt")
# Returns a data frame with columns:
#   rsID, chr_name, chr_position, effect_allele, other_allele, effect_weight

Calculating PRS

Single model

result <- calculatePRS(pgs_model)
result$prs                # Named numeric vector of PRS values
result$unmatched_by_chr   # Count of unmatched variants per chromosome
result$unmatched_rsIDs    # rsIDs of unmatched variants per chromosome
result$excluded_chr_counts # Variants on non-processed chromosomes

Multiple models (efficient)

fitPRSmodels() reads each chromosome's VCF data only once across all models:

model_files <- c("PGS000001.txt", "PGS000002.txt", "PGS000003.txt")
results <- fitPRSmodels(model_files)
results$prs  # Matrix: rows = samples, columns = models

Parallel processing

# Single model with parallel chromosome processing
result <- calculatePRS(pgs_model, parallel = TRUE, n_cores = 4)

# Multiple models in parallel
results <- fitPRSmodels(model_files, parallel = TRUE, n_cores = 4)

Lower-Level Functions

For more control, you can use the underlying functions directly:

# Summarize a single VCF file
vcf_summary <- summarizevcf("chr21.vcf.gz", genome = "hg38")

# Read dosage values for specific SNP indices
dosages <- getvcfsnp(vcf_summary, snp_indices = c(1, 5, 10))

# Match PGS model variants to VCF variants
matches <- findSNPs(pgs_model, vcf_summary$snp_summary$snp_df)

# Calculate PRS separately for REF and ALT matches
prs_alt <- calculatePRSalt(matches$alt_matches, pgs_model, vcf_summary)
prs_ref <- calculatePRSref(matches$ref_matches, pgs_model, vcf_summary)

How It Works

PRSfromPGS matches PGS model variants to VCF variants by genomic position and allele, then computes weighted dosage sums:

• ALT matches: When the effect allele matches the ALT allele, the VCF dosage (DS field) is used directly
• REF matches: When the effect allele matches the REF allele, the effect allele dosage is 2 - DS

The final PRS for each sample is the sum of effect_weight * effect_allele_dosage across all matched variants and all chromosomes.

Dependencies

Package	Source	Required
VariantAnnotation	Bioconductor	Yes
Rsamtools	Bioconductor	Yes
data.table	CRAN	Yes
BiocParallel	Bioconductor	No (parallel VCF indexing)
future	CRAN	No (parallel chromosome processing)
future.apply	CRAN	No (parallel chromosome processing)
parallelly	CRAN	No (parallel chromosome processing)

License

MIT