SUMMIT

Summary-statistics-based Unified Method for Multivariate Inference of Traits

SUMMIT estimates heritability and genetic correlation from GWAS summary statistics, computes LD scores from reference genotypes, and fits gene–environment interaction models and polygenic scores.

Main features

• Genome-wide and windowed LD-score estimation.
• SNP heritability and genetic correlation, including annotation partitions.
• h² and rg in cohorts with related individuals, keeping everyone and excluding related pairs.
• Batch analysis of many traits and annotation models.
• Gene–environment interaction (G×E) and environment-dependent residual variance.
• Joint models of multiple continuous or categorical environments.
• Cross-trait genetic response covariance through the research Python API.
• Additive and generalized G×E estimation for binary traits with PCGC.
• Quantitative epistasis with supplied targets and genetic backgrounds.
• G×E polygenic score fitting, scoring, and calibration.

Genotype input can be PLINK BED hard calls or biallelic diploid PGEN dosages. See the user guide for the inputs and availability of each method.

Install

You need Conda, Python 3.10 or newer, and a C++17 compiler with OpenMP support. The environment file supplies the Python and numerical-library dependencies, including OpenBLAS headers. Linux and macOS (Apple Silicon and Intel) support the optimized native G×E backend with BED input. On macOS, first install the Xcode Command Line Tools if needed with xcode-select --install.

git clone https://github.com/sriramlab/SUMMIT.git
cd SUMMIT
conda env create -f environment.yml
conda activate summit

On macOS, add OpenMP before building:

conda install -c conda-forge llvm-openmp

Then install SUMMIT on either platform:

CMAKE_ARGS=-DBLA_VENDOR=OpenBLAS python -m pip install .
summit --help
summit pgs --help
summit reference --help

See Installation for the complete setup instructions, development installs, and platform limits. If a build reports missing OpenBLAS headers, install the openblas development package as described in Troubleshooting.

Try it

Generate a small synthetic dataset and estimate partitioned LD scores:

python example/prepare_example_inputs.py
bash example/estimate_partitioned_gwldscore.sh
bash example/h2_ldscore.sh
bash example/rg_supplied_overlap_covariance.sh

Examples write to example/out/. They illustrate the commands; their small sample sizes are unsuitable for evaluating statistical performance. See the example guide for expected results, measured runtime, and G×E examples, including an explicit native-backend command that also runs on macOS.

Documentation

Commands and options lists the commands and shared analysis options.

• Input files
• LD scores
• Heritability and genetic correlation
• Related individuals
• Batch analyses
• G×E models
• Multiple environments
• Cross-trait response models
• Binary traits and PCGC
• Generalized G×E PCGC
• Quantitative epistasis
• G×E polygenic scores
• Real-data results
• Benchmarks

Manuscript reproduction

The manuscript package contains the aggregate source data and scripts for Figures 1–8 and S1–S44, supplementary metadata, and phenotype simulation utilities. Its plotting workflow uses relative paths and requires no individual-level data. See the installation notes for tested environments, hardware requirements, and installation time.

For a reproducible software reference, record the repository commit (git rev-parse HEAD) together with the environment used. The package version alone does not identify a particular source revision.

License

SUMMIT software is distributed under the MIT License. External datasets remain subject to their original access and use terms.

Citation

Jeong, M., Pazokitoroudi, A., Liu, Z., & Sankararaman, S. (2024). Scalable summary statistics-based heritability estimation method with individual genotype level accuracy. Genome Research, gr.279207.124. Paper.

Jeong, M., Liu, Z., Ni, J., An, U., Han, S., Anand, A., Anand, P., Boffelli, M., & Sankararaman, S. (2026). Genome-wide LD scores enable robust heritability and genetic correlation estimation from summary statistics across genetic ancestries. bioRxiv, 2026.10.20.760508. Preprint.