This repository contains the code necessary to reproducibly generate and analyze the simulated data described in (link submission).
All simulated data were generated using the GEMMR package, and all PLS analyses were performed using a modified version of pyls.
- Environment setup
The packages required and installation instructions can be found in ``requirements.txt`.
- Simulating data
In generate_data, any of the simulate_X.py files can be run with:
python simulate_X.py /path/to/outputs
The output directory will contain 1000 simulated X and Y matrices, each in a separate directory. The directory structure will have 10 "step" folders (e.g., one per sample size), and 100 "draw" folders inside each (i.e., one for each dataset within a given sample size).
- Running PLS
In pyls, run_pls_simulated.py can be run with:
python run_pls_simulated.py /path/to/single/dataset
This will create an output folder within the single dataset folder, containing various PLS outcome metrics (covariance explained, split-half stability, significance, etc.)
To speed up processing, a list of jobs can be created and submitted to a high-performance computing cluster.
- Analyzing results
From the analyze_results folder, a set of PLS results across 1000 datasets can be organized into single .pkl files using:
python organize_data.py /path/to/simulated/data
The output can then be analyzed, creating the plots seen in the manuscript, with:
python analyze_results.py /path/to/simulated_data
The scripts used to analyze the UK BioBank and PEPP datasets, which cannot be openly shared, are also provided for transparency.