FaST-LMM

FaST-LMM, which stands for Factored Spectrally Transformed Linear Mixed Models, is a program for performing genome-wide association studies (GWAS) on datasets of all sizes, up to one millions samples.

This release contains the following features, each illustrated with an IPython notebook.

• Core FaST-LMM (notebook) -- Lippert et al., Nature Methods 2011

Improvements:

• New features for single_snp (including effect size and multiple phenotype support) and epistasis (including reporting beta and using pre-computed eigenvalue decompositions) (notebook) -- Lippert et al., Nature Methods 2011
• Ludicrous-Speed GWAS (notebook) -- Kadie and Heckerman, bioRxiv 2018
• Heritability with Spatial Correction (notebook), Heckerman et al., PNAS 2016
• Two Kernels (notebook) -- Widmer et al., Scientific Reports 2014
• Set Analysis (notebook) -- Lippert et al., Bioinformatics 2014
• Epistasis (notebook) -- Lippert et al., Scientific Reports, 2013
• Prediction (notebook) -- Lippert et al., Nature Methods 2011

A C++ version, which is generally less functional, is available. See http://fastlmm.github.io/.

Quick install:

pip install fastlmm

FaST-LMM supports Python 3.10 through 3.14.

If you need support for BGEN files, instead do:

pip install fastlmm[bgen]

For best performance, be sure your Python distribution includes a fast version of NumPy. We use Anaconda's Miniconda.

Documentation

• IPython Notebooks:
  • Core, Epistasis, Set Analysis, Two Kernels
  • Multiple-Phenotype GWAS and related features
  • Heritability with Spatial Correction
  • Ludicrous-Speed GWAS
• Main Documentation
• Project Home and Full Annotated Bibliography

Code

• PyPi
• GitHub

Development

The project uses uv for reproducible development environments, dependency locking, testing, and builds:

uv sync --frozen --all-extras
cd tests
uv run --frozen --no-sync python test.py
cd ..
uv build --no-sources

See RELEASING.md for lower-bound, artifact, notebook, and release qualification.

Policy on AI-assisted development and contributions

AI tools may be used as productivity aids for drafting, exploration, and refactoring. Every contributed code or documentation change must be reviewed, edited, and validated by a human. AI does not replace design judgment, testing, or human responsibility for correctness.

The repository's published instructions and constraints for AI tools are in AGENTS.md.

Contacts

• Email the developers at fastlmm-dev@python.org.
• Join the user discussion and announcement list (or use web sign up).
• Open an issue on GitHub.