Installation

Requirements

• Python 3.10 or later

• pip

• PLINK2 on your PATH — only required if you plan to use --indep-pairwise LD pruning

PostHoc’s genotype I/O layer reads PLINK2 .pgen/.pvar/.psam filesets directly via pgenlib, so you do not need PLINK2 installed just to run commands — only for the optional LD-pruning QC step, which shells out to the plink2 binary.

Install from source

PostHoc is not yet published on PyPI. Install it directly from GitHub:

git clone https://github.com/Nifdi01/posthoc.git
cd posthoc
pip install -e .

This installs the core dependencies (PyTorch, NumPy, pandas, scikit-learn, Captum, statsmodels, Click) and exposes the posthoc command on your PATH.

Optional dependency sets

PostHoc defines two extras in pyproject.toml:

Extra	Installs	When you need it
genotype	pgenlib, pandas-plink	Always, in practice — pgenlib is what actually reads .pgen files. Install this unless you already have it.
dev	pytest, pytest-cov, black, ruff	Contributing to PostHoc or running the test suite.

Install either (or both) alongside the editable install:

# genotype I/O (recommended for any real usage)
pip install -e .[genotype]

# development tools
pip install -e .[dev]

# both
pip install -e ".[genotype,dev]"

Verifying the install

posthoc --help

should list the four subcommands: simulate-pheno, baseline, attribute, and pal.

GPU support

All model-training commands (attribute, pal) accept a --device option (default cpu). If you have a CUDA-capable GPU and a CUDA-enabled PyTorch build installed, pass --device cuda to train faster. PostHoc does not manage the CUDA/PyTorch install for you — follow the official PyTorch install instructions for your platform first.

Next steps

Continue to Quickstart for an end-to-end example, or Input data formats for details on the expected genotype/phenotype/covariate file formats.