pgscorer

An R package for computing Polygenic Risk Scores (PRS) from PGS Catalog scoring files against a BGZF-compressed VCF file.

The package unions SNP positions across all models, queries the VCF once per position in batches, and scores every model from the shared dosage table. On Windows batches run in-process; on Linux/macOS each batch runs in an isolated Rscript subprocess to keep peak memory bounded to one batch at a time.

Installation

remotes::install_github("USCbiostats/pgscorer")

Dependencies

• tabixr (Bioconductor) — BGZF/Tabix VCF queries
• data.table — fast PGS file reading

Usage

library(pgscorer)

results <- compute_prs(
  vcf_path   = "path/to/genotypes.vcf.gz",
  pgs_files  = c("PGS002164_hmPOS_GRCh37.txt.gz",
                 "PGS002863_hmPOS_GRCh37.txt.gz"),
  batch_size = 10000L,
  output_dir = "."       # writes one .rds per model; NULL to skip
)

If pgs_files is omitted, all files matching ^PGS.*\.txt\.gz$ in the current directory (or pgs_dir) are used automatically.

compute_prs() returns a named list of numeric vectors — one element per model, named by sample identifier.

Arguments

Argument	Default	Description
vcf_path	—	Path to .vcf.gz file (.tbi index must exist alongside it)
pgs_files	NULL	Character vector of PGS Catalog scoring file paths; NULL = auto-discover in pgs_dir
pgs_dir	"."	Directory searched for PGS files when pgs_files is NULL
chrom	NULL	Chromosome label (e.g. "21"); NULL = auto-detect from the .tbi index
batch_size	10000L	Positions per VCF query batch
output_dir	"."	Directory for output .rds files; NULL = do not save
verbose	TRUE	Print progress and summary

Output files

When output_dir is not NULL, one RDS file is written per model:

pgs_chr<chrom>_<model>_prs_combined.rds

Each file contains a named numeric vector of PRS values indexed by sample ID.

PGS Catalog file format

Scoring files should be downloaded directly from pgscatalog.org in the standard tab-delimited .txt.gz format. Harmonised position columns (hm_chr, hm_pos) are preferred over raw columns (chr_name, chr_position) when both are present.

VCF requirements

• BGZF-compressed (.vcf.gz) with a Tabix index (.vcf.gz.tbi)
• Must contain a DS (dosage) field in the FORMAT column

Allele effect formula

For SNPs where the effect allele matches REF:

PRS contribution = weight × (2 − dosage)

For SNPs where the effect allele matches ALT:

PRS contribution = weight × dosage

Rows where the effect allele matches neither REF nor ALT are excluded with a warning.

License

MIT © John Morrison