Reproduce the PYPE paper

Reproduction scope

The MR analyses use public summary statistics and can be reproduced automatically. The participant-level PheWAS requires the original authorized study data.

Public reproduction

Abdomen and liver aging variants, four outcome GWAS, all implemented MR methods, and Table 1 IVW comparisons.

Study-data reproduction

The full PheWAS, 15,082 tests, Table S20, and the main PheWAS figures require the study cohort.

Run the public analysis

Run these commands from a GitHub source checkout of the repository:

python -m pip install pype-mr openpyxl
python reproducibility/reproduce_paper.py \
  --output paper_reproduction \
  --bootstrap-iterations 1000 \
  --simulation-count 1000 \
  --seed 0

The first run streams about 2 GB of compressed public GWAS files and saves only the selected rows. Later runs reuse the extracted inputs.

Public sources

Public data sources used by the reproduction script.
Input	Source
Abdomen and liver aging variants	Le Goallec et al. Supplementary Data 1
Glycated haemoglobin	Neale Lab field 30750 raw GWAS
Body mass index	Neale Lab field 21001 raw GWAS
Glucose	Neale Lab field 30740 raw GWAS
Waist circumference	Neale Lab field 48 raw GWAS

Outputs

Files created by `reproducibility/reproduce_paper.py`.
File	Contents
`mr_results.tsv`	Current results for every method and exposure-outcome pair.
`ivw_comparison.tsv`	Published Table 1 values, legacy PYPE IVW values, and corrected current values.
`inputs/`	Small extracted public summary-statistic subsets.

Table 1 comparison

All eight legacy calculations match the published values within the displayed precision. The current IVW standard error follows the TwoSampleMR under-dispersion correction.

Published IVW values and results from the corrected current implementation.
Predictor	Outcome	Published beta	Published SE	Published p	Current p	Current SE
Abdomen	HbA1C	0.0265	0.0204	0.194	0.666	0.0614
Abdomen	BMI	0.0061	0.1158	0.958	0.957	0.1158
Abdomen	Glucose	-0.0151	0.0132	0.254	0.254	0.0132
Abdomen	Waist circumference	0.3556	0.1086	0.0011	0.0015	0.1121
Liver	HbA1C	0.0153	0.0441	0.729	0.730	0.0441
Liver	BMI	-0.0122	0.1231	0.921	0.921	0.1229
Liver	Glucose	0.0066	0.0072	0.358	0.380	0.0075
Liver	Waist circumference	0.2509	0.1570	0.110	0.110	0.1570

Why some current p-values differ

The original IVW implementation used the raw statsmodels residual-scaled standard error. TwoSampleMR prevents under-dispersion from making an IVW standard error smaller than the fixed-effect value. The current PYPE implementation applies that correction.

Effect estimates remain the same. Differences appear in standard errors and p-values only when the fitted residual scale is below one.

Bootstrap methods

Median, mode, and MR-PRESSO uncertainty estimates use random sampling. Fix the seed and number of bootstrap or empirical distributions when comparing runs. The original paper did not record a common seed for every table, so point estimates are the stable comparison target.