PenPRS: Penalized Regression for Inference of Polygenic Risk Scores

PenPRS is a Python package that includes summary statistics based sparse and accurate penalized regression models for the inference of Polygenic Risk Scores (PRS) using Linkage Disequilibrium (LD) and GWAS Summary Statistics. The models currently available include:

• LASSO
• SSL (Spike-and-slab LASSO PRS)
  • SSLAlpha (Alpha-heritability modeled SSL)

PenPRS integrates with magenpy for data harmonization and LD loading and computation. Details related to model methodology and specifications are present in the paper.

Reference

Song, J., Zabad, S., Yang, A., Gravel, S., & Li, Y. (2025). Sparse polygenic risk score inference with the spike-and-slab LASSO. Bioinformatics. https://doi.org/10.1093/bioinformatics/btaf578

Installation

To install the package from PyPI, use the following command:

pip install penprs

Quickstart

import magenpy as mgp
from penprs.model.Lasso import Lasso
from penprs.model.SSL import SSL
from penprs.model.SSLAlpha import SSLAlpha

# Load a genotype reference and GWAS summary statistics data (chromosome 22)
gdl = mgp.GWADataLoader(mgp.tgp_eur_data_path(),
                        sumstats_files=mgp.ukb_height_sumstats_path(),
                        sumstats_format="fastgwa")

# Compute block LD matrix:
ld_block_url = "https://bitbucket.org/nygcresearch/ldetect-data/raw/ac125e47bf7ff3e90be31f278a7b6a61daaba0dc/EUR/fourier_ls-all.bed"
gdl.compute_ld('block',
               ld_blocks_file=ld_block_url,
               dtype='int16',
               compute_spectral_properties=True,
               output_dir='temp/block_ld/')

# Initialize respective models
lasso_model = Lasso(gdl, lam=100)
ssl_model = SSL(gdl, l0=750, l1=10)

# Perform single fit using model initialized arguments
lasso_model.fit()   # fit using lam=100
ssl_model.fit()     # fit using l0=750, l1=10

Viewing model effect size estimates (BETA):

# Viewing the effect size estimates (BETA) for LASSO
lasso_model.to_table().head()

CHR	SNP	POS	A1	A2	BETA
22	rs131538	16871137	A	G	0.001439
22	rs9605903	17054720	C	T	-0.000000
22	rs5746647	17057138	G	T	-0.001534
22	rs16980739	17058616	T	C	-0.001342
22	rs9605923	17065079	A	T	0.006215

Pathwise and Warm-Start fashioned fit

# To solve in a path-wise or warm-start fashion
# save_intermediate=True to save effect size estimate per ladder step
lasso_model_pw = Lasso(gdl)
ssl_model_ws = SSL(gdl)

# Performs path-wise fit across the default 20 step ladder
lasso_model_pw.pathwise_fit(save_intermediate=True) 

# Performs warm-start fit across the default 20 step ladder
ssl_model_ws.warm_start_fit(save_intermediate=True)  

Command Line Interface (CLI)

You can fit the SSL model with Grid-Search (GS) from the command line using penprs_fit:

penprs_fit -l path_to_train_ld \
		-s path_to_train_sumstats \
		--output-dir path_to_output_dir \
		-m SSL \              # model: options = LASSO, SSL, SSLAlpha
		--hyp-search GS \     # hyperparameter search: options = GS (Grid Search), WS (Warm Start)
		--validation-ld-panel path_to_validation_ld \
		--validation-sumstats path_to_validation_sumstats \
		--use-symmetric-ld \
		--grid-metric pseudo_validation \
		--threads 4

For more information on the command line arguments, see penprs_fit --help

Citation

If you use this package in your research, please cite the following paper:

@article{10.1093/bioinformatics/btaf578,
    author = {Song, Junyi and Zabad, Shadi and Yang, Archer and Gravel, Simon and Li, Yue},
    title = {Sparse Polygenic Risk Score Inference with the Spike-and-Slab LASSO},
    journal = {Bioinformatics},
    pages = {btaf578},
    year = {2025},
    month = {10},
    doi = {10.1093/bioinformatics/btaf578},
    url = {https://doi.org/10.1093/bioinformatics/btaf578},
    eprint = {https://academic.oup.com/bioinformatics/advance-article-pdf/doi/10.1093/bioinformatics/btaf578/64738987/btaf578.pdf},
}