gbayes software validation

Current fitting uses gbayes::gbayes() with checked ordinary summary data and gbase Glist resources.

These recorded simulation results assess selected behavior on ten fixed datasets from the shared design. They do not revalidate the statistical method or cover every setting. The use snippets show the current package interfaces. This report step summarizes retained artifacts and does not rerun analyses. The gbayes guide describes current inputs and options.

Use

fit <- gbayes::gbayes(statistics$A, LD_Glist, method = "bayesc", trait = "A",
              prior = prior, control = list(residual_policy = "fixed",
                                          burnin = 500, sampling_sweeps = 700,
                                          seeds = c(11, 29, 47, 71)))

Results

Method	h2 estimate	h2 truth	bias	RMSE	intervals	effect r	max R-hat
bayesc	0.313	0.300	0.013	0.014	10/10	0.510	1.002
bayesr	0.306	0.300	0.006	0.010	10/10	0.513	1.000

The 50K-marker connected fits hold residual variance, effect scale and mixture weights fixed at declared values. They estimate marker effects and their posterior genetic summaries. Separate evidence covers parameter learning with a matched independent LD reference. The interval counts above describe conditional posterior intervals, not calibrated frequentist coverage.

Effect recovery · Posterior diagnostics

Block-eigen execution

The bounded block-eigen validation uses 512 markers, verified independent LD blocks and summary effects generated from the same reference. It confirms one native block load and eigensystem decomposition per block, square peak dense-block storage, and the exact full-positive-rank reduction to ordinary BayesR under matched controls.

For this synthetic spectrum, 99.5% retained positive mass kept 504 of 512 coordinates, so retained-factor memory fell by only 1.6%. Its LD-weighted posterior difference from the full fit was within the independent-seed Monte Carlo baseline. These results validate execution and diagnostics only; they do not establish a generally optimal retention threshold, population-scale speed, or external SBayesRC equivalence.