Standalone mixed models with YAML

Run a model, find its results, or build and validate an installation.

Use one YAML file for input data, the model, covariance matrices, threads and outputs. The same command runs supplied-covariance solving and REML estimation.

Run a model

On GenomeDK, start inside a Slurm compute allocation with enough CPUs for the threads requested in your YAML. Use your project’s shared installation:

export PATH="/path/to/shared/installation/bin:$PATH"
gsuite run model.yml

No compilation or access to the maintainer’s Conda environment is needed. Input and output paths resolve beside the YAML. Choose a fresh output directory; scratch processing and copying results back are automatic.

To check or inspect a model before fitting:

gsuite validate model.yml
gsuite describe model.yml

See the complete ssSNPBLUP example for input files, covariance matrices, thread settings and outputs.

Find the results

Results are saved to output.directory in the YAML:

  • Prediction CSVs and requested effect tables contain the estimates.
  • run.json records numerical acceptance and diagnostics; execution.json reports workflow timings and memory.
  • Optional truth comparison writes accuracy, bias and calibration to evaluation/summary.csv.

Check acceptance before using estimates. See results and diagnostics for file definitions, failure handling and resource reporting.

Build, validate and distribute

Maintainers: inside an extracted source bundle on a compute node, with the build environment activated, run each task separately after the previous one succeeds:

bash build.sh

Builds and installs, reusing compatible native cache entries.

bash hpc_tasks/validate.sh

Runs exact references, truth-evaluation checks and all 20 fixture models.

bash hpc_tasks/build_binary_distribution.sh

Produces a binary archive and validates a fresh installation. Logs and results are collected automatically. See installation and maintenance for prerequisites, acceptance checks and binary installation instructions.

Reference