ExPRSweb: An online repository with polygenic risk scores for common health-related exposures

Ying Ma, Snehal Patil, Xiang Zhou, Bhramar Mukherjee, Lars G. Fritsche

TL;DR

Researchers have developed ExPRSweb, an online repository featuring over 300 Polygenic Risk Scores (PRS) specifically for health-related exposures (like BMI, smoking, and vitamin levels). By using these genetic proxies, the team demonstrated that combining multiple exposure-based scores can significantly improve our ability to predict chronic diseases compared to using disease-focused genetic scores alone.

What Problem Does This Paper Address?

In the world of precision medicine, we often look at two things: your DNA (genetics) and what you do/where you live (exposures). While we are getting better at calculating a Polygenic Risk Score (PRS)—a single number summarizing an individual's genetic liability for a disease—measuring "exposures" is notoriously difficult.

Electronic Health Records (EHRs) are often missing data on lifestyle factors like smoking or diet. Even when the data exists, it’s prone to "measurement error" and "recall bias" (people aren't always honest about how much they smoke or weigh). This paper introduces Exposure PRSs (ExPRSs). Instead of measuring the exposure itself, researchers use genetic variants associated with that exposure as a stable, birth-assigned proxy. This allows scientists to study the "genetic pull" toward certain behaviors or physiological states even when clinical data is missing.

Methods

The authors built a uniform analytic framework to generate and validate ExPRSs for 28 common exposures.

  • The Algorithms: They didn't just pick one method; they compared four popular PRS construction techniques:
  • C+T (Clumping and Thresholding): The classic approach of picking the most significant, independent genetic markers.
  • Lassosum: A penalized regression method.
  • DBSLMM: A deterministic Bayesian approach.
  • PRS-CS: A Bayesian method using "continuous shrinkage" priors.
  • Validation: They used a "split-sample" approach, training the models on 50% of their data and testing on the other 50% to ensure the scores weren't just "memorizing" the noise in the data.
  • Innovation: They introduced "multiExPRS," a model that combines multiple different exposure scores to predict a single chronic condition, and "Exclusion-PRS-PheWAS," a technique to see if genetic risk for an exposure still predicts disease even in people who don't actually manifest that exposure (e.g., does a high-BMI genetic score still predict heart disease in thin people?).

Datasets

The study utilized two massive, well-known biobanks:

  • Michigan Genomics Initiative (MGI): A hospital-based cohort of ~46,782 unrelated individuals of European ancestry.
  • UK Biobank (UKB): A population-based cohort of ~408,595 individuals.

The researchers used these to evaluate 82 sets of GWAS (Genome-Wide Association Study) summary statistics. They ultimately established high-performing ExPRSs for 27 distinct exposures.

Key Insights and Ideas

  • No "One Size Fits All": One of the most important takeaways is that different PRS methods work better for different traits. For example, PRS-CS was the best for 18 out of 24 exposures in the MGI cohort, but Lassosum excelled for alcohol and smoking, while DBSLMM was superior for lipid levels.
  • Genetic Proxies as Time Machines: The "Exclusion-PheWAS" results were fascinating. They found that the BMI ExPRS was strongly associated with bariatric surgery even in patients who currently had a "healthy" BMI. This suggests the ExPRS captures a "pre-treatment" state or a lifelong genetic predisposition that clinical measurements taken at a single point in time might miss.
  • The Power of Multiplicity: Traditionally, if you want to predict heart disease, you use a heart disease PRS. This paper shows that adding ExPRSs for BMI, smoking, and cholesterol on top of the disease-specific score provides a "substantial improvement" in prediction for several chronic conditions.

Key Results

  • Repository Success: The authors created ExPRSweb, an interactive platform to browse performance metrics for 336 promising ExPRSs.
  • Quantitative Gains: Adding multiple ExPRSs to a disease-specific PRS (YPRS) improved the Area Under the Covariate-adjusted ROC Curve (AAUC) for several conditions. For instance, the AAUC for predicting hypertension increased from 0.627 to 0.637 when adding a suite of ExPRSs.
  • Risk Stratification: The "multiExPRS" approach was particularly effective at identifying high-risk individuals. For heart failure, the odds ratio (OR) for the top 1% of the risk distribution jumped from 1.13 (using only the disease PRS) to 2.72 (when adding exposure scores).

Limitations and Future Work

  • Ancestry Bias: The study focused almost exclusively on individuals of European ancestry due to the limitations of current biobanks. The authors explicitly state that these scores may not transfer well to other ancestry groups, highlighting a critical need for more diverse genomic data.
  • Genetic vs. Actual Environment: An ExPRS only captures the heritable fraction of an exposure. It cannot account for the actual environment or sudden lifestyle changes. It is a measure of "predisposition," not a perfect mirror of reality.
  • Future Directions: The authors suggest that ExPRSs could be used to "impute" missing exposure data in EHRs, which could be a game-changer for large-scale epidemiological studies where clinical data is patchy.

Web Resource: You can explore the findings and scores at ExPRSweb.

Summary updated

Abstract

Abstract Complex traits are influenced by genetic risk factors, lifestyle, and environmental variables, so called exposures. Some exposures, e.g., smoking or lipid levels, have common genetic modifiers identified in genome-wide association studies. Since measurements are often unfeasible, Exposure Polygenic Risk Scores (ExPRSs) offer an alternative to study the influence of exposures on various phenotypes. Here, we collected publicly available summary statistics for 28 exposures and applied four common PRS methods to generate ExPRSs in two large biobanks, the Michigan Genomics Initiative and the UK Biobank. We established ExPRS for 27 exposures and demonstrated their applicability in phenome-wide association studies and as predictors for common chronic conditions. Especially, the addition of multiple ExPRSs showed, for several chronic conditions, an improvement compared prediction models that only included traditional, disease-focused PRSs. To facilitate follow-up studies, we share all ExPRS constructs and generated results via an online repository called ExPRSweb.
BibTeX
@article{ma2022exprsweb,
  author = {Ying Ma and Snehal Patil and Xiang Zhou and Bhramar Mukherjee and Lars G. Fritsche},
  title = {{ExPRSweb: An online repository with polygenic risk scores for common health-related exposures}},
  journal = {The American Journal of Human Genetics},
  year = {2022},
  doi = {10.1016/j.ajhg.2022.09.001},
}

Loading related content...

Figure