prsedm 1.1.0

PRSedm (Polygenic Risk Score Extension for Diabetes Mellitus) is a package for local and remote generation of Polygenic Risk Scores (PRS) for Diabetes Mellitus (DM)

PRSedm

Overview

PRSedm (Polygenic Risk Score Extension for Diabetes Mellitus) is a flexible and extendable open-source package for efficient local and remote (All of Us, UK Biobank, etc.) generation of published Polygenic Risk Scores (PRS) for Diabetes Mellitus (DM) and related cardiometabolic phenotypes.

PRS for Type 1 diabetes (T1D) and Type 2 diabetes (T2D), and more recent partitioned Polygenic Scores (pPS), have numerous applications as research and clinical tools.

PRSedm aims to introduce a new parallelized "one-liner" method to generate standardized PRS and pPS for DM robust to variables such as genotyping method, quality control, and imputation panel.

Installation

Dependencies

PRSedm requires the following packages:

• Python (>=3.9)
• Joblib (>=1.3.2)
• Pandas (>=2.2.3)
• Pysam (>=0.22.0)
• Numpy* (2.x/1.x)

*Build with 1.x when deploying to RAP platforms with 1.x dependencies.

User Installation

PRSedm is available through a number of channels:
PIP: pip install prsedm
Anaconda: conda install sethsh7::prsedm
Build from source: python -m build

Usage

PRS Database (variants.db).

Beginning with v1.1.0, the PRS-EDM variant database is no longer packaged due to size limits and is instead hosted on Zenodo: https://zenodo.org/records/17903390 (direct download: https://zenodo.org/records/17903390/files/variants.db?download=1). The CLI now supports --getsql, which prints the path to the database and downloads it automatically if missing. PRSedm will first check the environment variable PRSEDM_SQL_PATH, then any cached copy, and finally download from Zenodo if needed.

Command Line Interface

To call PRSedm from the command line:

prsedm --vcf <path_to_vcf_file> [options]

• --vcf (required): Path to an indexed VCF or BCF file, or a text file mapping one VCF per contig.
• --col: Genotype column to score (default: GT, options: GT for WGS, GP for imputed data).
• --build: Genome build to use (default: hg38, options: hg19, hg38).
• --scores (required): Comma-separated list of PRS to generate, e.g., PRS1,PRS2.
• --impute (optional): Enable imputation (requires --refvcf) (default: off).
• --refvcf (optional): Path to indexed reference VCF/BCF, or text file mapping one VCF/BCF per contig (required if --impute is set).
• --norm (optional): Perform fixed MinMax normalization (default: off).
• --parallel (optional): Enable parallel processing (default: off).
• --ntasks (optional): Number of tasks to use for parallel processing (default: CPU count).
• --batch-size (optional): Number of variants per batch (default: 1).
• --output: Path to save the output file (default: results.csv).

Python (recommended)

To call PRSedm from python:

import PRSedm
df = prsedm.gen_dm(vcf, col, build, scores, impute, refvcf, norm, parallel, ntasks, batch_size)

Single file per-chromosome loading

For --vcf and --refvcf you can point to a text a single file mapping per contiguous region formatted as such (whitespace delimited):

file1.chr1.vcf.gz   chr1
file2.chr2.vcf.gz   chr2
...

Research Analysis Platforms (RAP's)

Remote deployment to remote Research Analysis Platforms (RAP's) is possible via notebook wrappers:

• All of Us (WGS) - Notebook Here
• UK Biobank (imputed WGS)- Notebook Here
• UK Biobank (imputed array) - Notebook Here

Available PRS

PRS schematics are stored in a SQLlite database (variants.db) and accessed via a JSON metadata (prs_meta.json). These are automatically read and fully extendable if the user wishes to add additional PRS. The following PRS are available by default:

Type 1 Diabetes

Flag	Method	Variants	Description	PMID
t1dgrs2-luckett25	HLA Interaction + Partitioned	67	"GRS2x" updated PRS with widest compatibility and HLA-based risk pPS.	40267362
t1dgrs2-qu22	HLA Interaction + Partitioned	71	Original "GRS2" PRS with the addition of 4 African ancestry SNPs from Onengut, proposed in Qu et al and utilized in eMERGE.	34997821
t1dgrs2-sharp21	HLA Interaction + Partitioned	67	Version of "GRS2" PRS designed for "TOPMED-R2" from 2021 GitHub.	35312757
t1d-onengut19-afr	Additive	6	African-ancestry PRS proposed by Onengut in 2019, updated for modern compatibility.	30659077
t1dgrs2-sharp19	HLA Interaction + Partitioned	67	Original 1000 Genomes version of "GRS2" PRS as published, with limited modern compatibility.	30655379

Type 2 Diabetes

Flag	Method	Variants	Description	PMID
t2dp-suzuki24-ma	Additive + Partitioned	1289	Multiancestry weighted Suzuki T2D index variant PRS, and pPS from hard-clustering analyses.	38374256
t2dp-suzuki24-<ancestry>	Additive + Partitioned	1128 - 1285	As above but weighted for specific ancestries <eur/afr/safr/eas/sas/his>	38374256
t2dp-smith24-ma	Additive + Partitioned	353	Multiancestry cluster-weighted Smith T2D index variant PRS, and pPS from soft-clustering analyses.	38443691
t2dp-smith24-<ancestry>	Additive + Partitioned	25 - 490	As above but from ancestry-specific soft clustering <eur/afr/eas/amr>.	38443691
t2d-mahajan22-ma	Additive	338	Older PRS from Mahajan et al composed of multiancestry index variants.	35551307
t2dp-udler18	Additive + Partitioned	67	T2D pPS from first soft-clustering analysis.	30240442

Other

Flag	Phenotype	Method	Variants	Description	PMID
cdgrs-sharp25	Celiac Disease	HLA Interaction + Partitioned	42	Modernized Celiac disease PRS and pPS with similar model to "GRS2x", utilized for combined screening.	32790217

Additional Features

HLA Interaction PRS (+GRS2x Update)

PRSedm features a complete algorithm for GRS which incorporate HLA interaction terms as previously published by us such as T1D-GRS2 (or just GRS2). A number of advancements have been added to improve the generation of HLA interaction, described as GRS2x.

HLA Type Estimation and LD Tiebreak

HLA alleles can be estimated by proxy (or tag) single nucleotide polymorphisms alone and predictions are output e.g. (DR3-DQ2.5/DR3-DQ2.5) Due to imperfect proxy SNPs >2 HLA calls can be made in interaction scores such as GRS2, a probablistic tiebreaker algorithm using a HLA reference frequencies (Klitz et al) now resolves impossible numbers of calls without excluding any samples.

Missing variant mean effect imputation (optional)

PRSedm optionally uses Hardy-Weinberg Equilibrium with a reference VCF/BCF legend (ensure you have variant frequency coded as 'AF', genotypes not required) to impute the mean effect size for missing SNPs, handle missing variants, and enable static normalization.

• dbsnp hg38 - TOPMED Bravo Freeze 8, or NCBI ('AF' field added) are recommended.
• dbsnp hg19 - 1000 Genomes, Haplotype Reference Consortium, NCBI GRCh37 recommended.

Minimum and Maximum Normalization (optional)

PRSedm hardcodes static normalization of minimum and maximum potential risk contribution (no risk alleles vs all risk alleles) creating a scale of 0-1. Static normalization with imputation ensures that PRS values translate to a common relative risk scale across datasets. Imputation must be enabled, or all variants must be present.

Development

Developed and maintained by Seth A. Sharp ([e-mail]) at the Translational Genomics of Diabetes, Stanford University, with collaboration from colleagues at the University of Exeter and MGH/Broad Institute. Lu Zhang and Han Sun contributed to v1.1.0 onwards.

License

This project is licensed under the MIT License (Non-Commercial).

• Academic, research, and personal use are allowed.
• Commercial use is prohibited without prior permission.

See the LICENSE file for full details.