Software validation

Focused simulation checks, numerical evidence and limits for selected gsuite implementations.

These focused simulation checks ask whether selected gsuite implementations behave as expected with known truth or independent numerical calculations. The statistical methods were established in the original literature; these checks do not establish them anew or cover every supported setting. Tutorials teach workflows, while Benchmarks compare performance.

Shared genomic simulation

The shared simulation design (gsim) fixes ten seeds, 10,000 individuals, 50,000 markers and known generating effects. Each page shows how to use one library and what was checked on the same datasets:

• glma: association scans
• gcorr: heritability and genetic correlation
• gbayes: Bayesian regression
• gscore: held-out prediction
• gmap: regional mapping and colocalisation
• gsea: genes and pathways

The worked analysis connects the six modules in one workflow. Results are descriptive for the specified simulation; the library pages state the relevant limits.

Other evidence

• Standalone GenomeDK installation: recorded binary acceptance, artifact hashes and limits of the installer checks.
• Earlier connected genomic checks: prior focused checks, including a separate reference-LD parameter-learning qualification.
• gcorr method comparison: six summary-statistic methods on a separate two-trait simulation, with figures, replicated errors and availability.
• Current scope and qualification limits: package-level and model-level checks that remain incomplete.

Evidence from different simulation designs is identified separately and is not pooled.