Python pipeline for PheWAS and Mendelian randomization

PYPE runs phenome-wide association studies (PheWAS), creates PheWAS plots, and performs Mendelian randomization (MR) in Python.

User guide | Roadmap | Reference paper

Features

• PheWAS with genotype or phenotype predictors
• Covariate-adjusted linear regression
• Bonferroni, Sidak, and Benjamini-Hochberg correction
• Manhattan, volcano, and category enrichment plots
• Variant-to-gene mapping and optional BioThings annotations
• Eleven MR methods, including MR-Egger, weighted median, weighted mode, and MR-PRESSO

Installation

python -m pip install pype-mr

Optional annotation dependencies:

python -m pip install "pype-mr[annotations]"

Reproduce the paper

The public Mendelian randomization analyses can be rerun from a GitHub source checkout using the Le Goallec and Neale Lab summary statistics:

python -m pip install -e . openpyxl
python reproducibility/reproduce_paper.py

See reproducibility/README.md for the data boundary and expected outputs.

PheWAS

Pass one dataframe containing phenotypes and covariates, and another containing the predictors. Both dataframes must have the same sample index.

import pandas as pd
import pype

phenotypes = pd.DataFrame({
    "trait_a": [1.2, 2.0, 1.5, 3.1],
    "trait_b": [0.0, 1.0, 0.0, 1.0],
    "age": [50, 61, 47, 70],
})

predictors = pd.DataFrame({
    "variant_1": [0, 1, 1, 2],
})

results = pype.phenome_wide_association(
    phenotypes,
    predictors,
    outcomes=["trait_a", "trait_b"],
    covariates=["age"],
    min_sample_count=3,
)

The result uses consistent snake_case columns: outcome, predictor, sample_count, total_sample_count, negative_log10_pvalue, pvalue, beta, and standard_error.

Plotting

Plots use the PheWAS result dataframe plus a small metadata table containing phenotype descriptions and categories.

import pandas as pd
import pype
from pype.plotting import category_enrichment, manhattan, volcano

metadata = pd.DataFrame({
    "outcome": ["trait_a", "trait_b"],
    "description": ["Trait A", "Trait B"],
    "category": ["Measurements", "Diagnoses"],
})
results = pype.add_phenotype_metadata(results, metadata)

manhattan(results, "manhattan.png")
category_enrichment(results, "category_enrichment.png")
volcano(results, "volcano.png")

Mendelian randomization

Exposure and outcome dataframes must contain:

import pandas as pd
import pype

exposure = pd.read_csv("exposure.tsv", sep="\t")
outcome = pd.read_csv("outcome.tsv", sep="\t")

results, diagnostics = pype.mendelian_randomization(
    exposure,
    outcome,
    exposure_name="exposure_trait",
    outcome_name="outcome_trait",
    methods=("ivw", "egger", "weighted_median"),
    seed=0,
)

Use methods="all" to run every available method. PYPE aligns effect alleles, flips reversed effects, and removes incompatible or ambiguous palindromic variants before analysis.

Available methods:

• Inverse variance weighted
• MR-Egger
• Simple, weighted, and penalized weighted median
• Simple, weighted, penalized, and NOME mode estimators
• MR-PRESSO

Data

Tests use synthetic data. Study data are supplied by users under the access terms of their data source.

License

Apache License 2.0. See LICENSE.md.

Citation

@article{Dalal2024PYPE,
  title   = {PYPE: A pipeline for phenome-wide association and Mendelian randomization in investigator-driven biobank scale analysis},
  author  = {Dalal, Taykhoom and Patel, Chirag J.},
  journal = {Patterns},
  year    = {2024},
  volume  = {5},
  number  = {6},
  pages   = {100982},
  doi     = {10.1016/j.patter.2024.100982},
  url     = {https://doi.org/10.1016/j.patter.2024.100982}
}