PolyGenie: a reproducible Nextflow pipeline for phenome-wide association studies using polygenic risk scores

PolyGenie is an open-source, reproducible Nextflow pipeline that standardizes the execution and interactive visualization of polygenic risk score-based phenome-wide association studies (PheWAS) across diverse population cohorts.

The Gist

The Big Picture: A "Universal Translator" for Genetic Risk

Imagine you have a massive library of books (your DNA) that tell a story about your health. Scientists have figured out how to read specific chapters of these books to predict your risk for certain diseases. These predictions are called Polygenic Risk Scores (PRS). Think of a PRS like a "credit score" for your health, but instead of money, it measures your genetic likelihood of getting things like heart disease, diabetes, or depression.

The problem? Until now, using these "credit scores" to check your risk for everything at once was a nightmare. It was like trying to build a house by hand, brick by brick, for every single new neighborhood. Every time a researcher wanted to test a new group of people, they had to write new code, fix broken tools, and struggle with messy data.

PolyGenie is the solution. It is a digital construction crew (a software pipeline) that automates the whole process. It takes your genetic "credit scores" and your health records, runs a massive check-up, and builds a beautiful, interactive dashboard so anyone can see the results.

---

How It Works: The Assembly Line

Think of PolyGenie as a highly efficient, automated factory line with three main stations:

1. The Intake Station (Input)

You don't need to be a computer wizard to use this. You just bring two things to the factory:

• The Scores: Your pre-calculated genetic risk scores (like a list of numbers).
• The Health Data: Your medical records, lifestyle habits, and blood test results.
• The Blueprint: A simple instruction sheet (a configuration file) that tells the machine where to find the data.

Analogy: It's like dropping your car off at a mechanic. You don't need to know how to fix the engine; you just hand them the keys and say, "Check the brakes and the oil."

2. The Processing Station (The Engine)

Once the data is in, PolyGenie goes to work. It uses a powerful tool called Nextflow, which is like a super-efficient traffic controller.

• It checks if the data is clean (no missing pieces).
• It runs thousands of mathematical tests simultaneously. It asks questions like: "Do people with high genetic risk for obesity also tend to have high blood pressure?" or "Do people with high risk for depression also have lower energy levels?"
• It does this for 135 different traits at once, comparing them against thousands of health outcomes.

Analogy: Imagine a detective who can interview 10,000 suspects in a single afternoon, cross-referencing their alibis with a massive database, and instantly finding patterns that would take a human team years to spot.

3. The Showroom (Visualization)

After the math is done, the results are stored in a digital filing cabinet (a database). But instead of giving you a boring spreadsheet, PolyGenie builds a video game-style dashboard (using a web app called Dash).

• The Map: You can see a giant scatter plot where every dot is a disease or trait. If a dot is high up, it means a strong genetic link.
• The Slider: You can slide a bar to see how risk changes as your genetic score goes from low to high.
• The Split: You can even see if the results are different for men and women.

Analogy: Instead of reading a 500-page report on the weather, you look at a live, interactive 3D map of the globe that shows rain, wind, and temperature in real-time. You can zoom in, click on a city, and see exactly what's happening.

---

Why This Matters: The "Frailty" Example

To prove it works, the researchers used PolyGenie on a real group of people (the GCAT cohort in Spain). They looked at Frailty (the risk of getting weak and frail as you age).

They asked: "If someone has a high genetic risk for frailty, what else are they at risk for?"

The dashboard instantly showed two clear patterns:

1. Obesity: As the frailty risk went up, the risk of being overweight went up too.
2. Depression: As the frailty risk went up, the risk of depression went up, but much more sharply for women than for men.

This kind of discovery—seeing how one genetic risk connects to multiple different health issues and affects men and women differently—happened in minutes, not months.

---

The Best Part: It's "Plug-and-Play"

Most scientific tools are like custom-made suits; they only fit the person they were made for. If you want to use them for a different group of people, you have to re-tailor the whole thing.

PolyGenie is like a "One-Size-Fits-All" smart suit.

• It is Open Source: Anyone can download it for free.
• It is Portable: It works on a laptop, a massive supercomputer, or in the cloud.
• It is Adaptable: If you have a new group of patients, you just swap out the data files. You don't have to rewrite the code.

In Summary

PolyGenie is a tool that takes the complex, scary world of genetic data and turns it into a clear, interactive story. It allows doctors and researchers to stop wrestling with code and start asking big questions: "How do our genes connect our different health problems?"

By making this process easy and reproducible, it helps us understand the hidden links between our DNA and our daily lives, paving the way for better, more personalized healthcare for everyone.

Technical Summary

1. Problem Statement

While Polygenic Risk Scores (PRS) are powerful tools for quantifying genetic susceptibility to complex traits, and Phenome-wide Association Studies (PheWAS) allow researchers to explore the shared genetic architecture across diverse phenotypes, there is a significant gap in standardized, portable tools for executing these analyses.

• Lack of Standardization: Existing tools are often fragmented. Static resources (e.g., PRS Atlas, PheWAS Catalog) are tied to specific cohorts and cannot be redeployed. Analytical tools (e.g., R PheWAS, pyPheWAS) require command-line expertise and lack integrated visualization. Visualization platforms (e.g., PheWAS-ME) often function as post-hoc layers requiring separate statistical analysis.
• Reproducibility and Scalability: Systematic execution of PheWAS across hundreds of PRS-phenotype combinations in arbitrary population cohorts is difficult to reproduce, scale, and adapt due to heterogeneous phenotype data and the need for custom coding for regression models and visualization.

2. Methodology

PolyGenie is an open-source, end-to-end pipeline designed to bridge the gap between PRS generation and downstream PheWAS analysis/visualization.

• Architecture & Implementation:

  • Built using Nextflow (DSL2) to ensure portability across local laptops, HPC clusters, and cloud environments.
  • Input: Accepts precomputed PRS (allowing users to use any PRS generation method like PRSice-2, PRScs, or MegaPRS) and cohort phenotype data.
  • Configuration: Uses a metadata-driven design with a single YAML configuration file and two CSV metadata files (one for PRS, one for phenotypes). This allows adaptation to new cohorts without modifying the pipeline code.
  • Data Handling: Supports binary (diseases) and continuous (metabolites, lifestyle) outcomes.

• Workflow Stages:

  1. Preprocessing:
     • CHECK_PRS_FILES: Validates file existence and column structure.
     • CHECK_PHENOTYPE_FILES: Validates phenotype variables and filters out binary traits with insufficient case counts (default <10 cases).
  2. Analysis:
     • COMPUTE_PRS_PERCENTILES: Calculates prevalence or mean values of phenotypes across PRS percentile bins (default 100 bins).
     • COMPUTE_PRS_REGRESSIONS: Performs linear (for continuous) or logistic (for binary) regression for every PRS-phenotype pair.
     • Parallelization: Runs analyses in parallel across PRS and phenotypes using Nextflow's executor abstraction and Python's joblib.
     • Adjustment: Models are adjusted for covariates (e.g., age, sex, principal components) specified globally or per-phenotype.
  3. Storage & Visualization:
     • Results are ingested into a SQLite database.
     • A Plotly Dash web application queries the database to render interactive visualizations.

• Case Study Implementation (GCAT Cohort):

  • Applied to the GCAT cohort (n≈5,000, Iberian descent).
  • Inputs: 135 precomputed PRS (using MegaPRS) and 1,483 phenotypes (ICD-10, metabolites, questionnaires).
  • Analysis: Over 200,000 associations tested. Models adjusted for age, sex, and 10 PCs. Bonferroni correction applied per phenotype category.

3. Key Contributions

• Decoupled Design: Separates PRS generation from the analysis pipeline, making it method-agnostic and applicable to any cohort with precomputed scores.
• End-to-End Automation: Automates the entire workflow from data validation to statistical regression and database storage.
• Interactive Visualization: Provides a user-friendly Dash interface featuring:
  • PheWAS Scatter Plots: Signed log₁₀(p-value) across phenotypic domains.
  • Prevalence-by-Percentile Plots: Visualizing dose-response relationships, stratified by sex.
  • Exportable Tables: Top hits and summary statistics.
• Portability: The Nextflow implementation ensures the pipeline runs seamlessly across different computing infrastructures.
• Reproducibility: Includes a fully worked example using the GCAT cohort, serving as a template for other researchers.

4. Results

• Technical Validation: The pipeline successfully processed 135 PRS against 1,483 phenotypes, generating over 200,000 associations and populating the SQLite database.
• Biological Insight (Frailty Example):
  • The authors demonstrated the tool's utility by analyzing the Fried frailty PRS against overweight/obesity (E66) and major depressive disorder (F32).
  • Findings:
    • A clear dose-response relationship was observed between higher frailty PRS and increased prevalence of both conditions.
    • Sex Differences: While obesity trends were similar across sexes, the prevalence of major depressive disorder was significantly higher in females across all PRS percentiles, aligning with known epidemiological data.
  • This confirmed the tool's ability to detect pleiotropic effects and sex-specific genetic risk patterns without custom coding.

5. Significance

• Bridging the Gap: PolyGenie fills a critical niche by combining the statistical rigor of command-line tools with the accessibility of interactive dashboards, eliminating the need for researchers to coordinate multiple specialized tools.
• Democratization of Genomics: By removing the need for programming expertise in the analysis and visualization phases, it enables researchers to conduct systematic PheWAS analyses and generate hypotheses rapidly.
• FAIR Compliance: Designed as a FAIR (Findable, Accessible, Interoperable, Reusable) resource, it supports long-term, sustainable genomic research across European cohorts.
• Future Directions: While currently limited to frequentist regression (lacking Mendelian randomization or colocalization), the modular design allows for future expansion into federated analyses and more complex causal inference methods.

Availability: The pipeline is open-source under the MIT License at github.com/gcatbiobank/polygenie-pipeline, with a live demo at polygenie.igtp.cat.