OmicsBraid 0.2.2
Frozen manuscript release
- Frozen the covariance-aware v0.2.2 statistical core used for
manuscript analyses.
- Provides subject-level permutation calibration for the cross-omic
omnibus test.
- Provides null-shift matched-subject bootstrap calibration for
Q_omics.
- Retains analytic confidence intervals as default reporting and BCa
layer intervals as sensitivity analysis.
- Retains covariance-aware GLS trajectory inference and hierarchical
confirmatory/suggestive braid classification.
- Includes Evidence Forest, Effect Braid, braid heatmap, concordance
map, export helpers, and known-truth simulation.
- Adds public-facing GitHub/pkgdown/release documentation without
altering the frozen
R/ statistical source.
Development history
OmicsBraid 0.2.2
Robust empirical calibration
- Adds
empirical_omics_tests() for resampling-calibrated
cross-omic inference.
- Global omnibus evidence can be calibrated by matched-subject
group-label permutation or by centered matched-subject bootstrap.
- Cross-omic heterogeneity can be calibrated by a raw-data null-shift
matched-subject bootstrap (recommended) or an effect-level centered
bootstrap under the fitted common-effect null; naive label permutation
is deliberately not used for this composite null.
- The permutation omnibus uses a covariance-aware Mahalanobis/Wald
statistic calibrated by the joint subject-level permutation
distribution. Empirical heterogeneity uses a covariance-aware GLS
residual quadratic statistic calibrated by matched-subject bootstrap
nulls. Both avoid requiring a chi-square reference distribution under
heavy tails.
- Asymptotic p-values are preserved alongside empirical p-values.
run_omics_braid() reports empirical tests only when
explicitly requested; empirical p-values are not made primary unless
empirical_use_as_primary = TRUE.
- Layer CI workflow now also exposes the already-supported
basic bootstrap interval.
- Adds a final targeted robust-calibration benchmark comparing
asymptotic, permutation, centered-bootstrap, and null-shift-bootstrap
Type-I error, inversion power, and analytic/basic/percentile/BCa
interval coverage.
- Adds persistent internal-disk checkpointing for the final
robust-calibration run (
04_RUN_ROBUST_CALIBRATION.R).
- The validated v0.1.9/v0.2.1 braid classifier logic is otherwise
unchanged.
OmicsBraid 0.2.1
- I/O-resilience patch for confirmatory validation; statistical
algorithms and simulation design are unchanged from v0.2.0.
- Confirmatory checkpoints and high-frequency outputs are now written
to persistent internal-disk storage under
~/OmicsBraid_ValidationCache/confirmatory_v020_design.
- Valid checkpoints from an interrupted v0.2.0 run are imported
automatically; incomplete/corrupt RDS files are ignored.
- Checkpoints are validated before reuse and written atomically via
temporary-file + rename.
- Final validation outputs are synchronized back to the package
_CONFIRMATORY_VALIDATION_OUTPUT folder only after the local
run completes.
OmicsBraid 0.2.0
- Froze the v0.1.9 omnibus/GLS/Q/equivalence/trend/classification
definitions for confirmatory validation rather than continuing
classifier redesign.
- Added
bootstrap_effect_intervals() with
subject-bootstrap percentile, basic, and BCa confidence intervals for
layer-specific Hedges’ g effects.
- Added
bootstrap_consensus_intervals() with
percentile/basic bootstrap confidence intervals for GLS consensus
effects.
- Added end-to-end
ci_method and
integrated_ci_method options to
run_omics_braid() while deliberately retaining analytic SEs
and p-values as the inferential basis.
- Preserved analytic intervals alongside bootstrap intervals
(
conf_low_analytic, conf_high_analytic) so
interval-method sensitivity is auditable.
- Added a targeted confirmatory simulation runner with n/group =
20/40/80/160/320, normal versus heavy-tailed residuals, Monte-Carlo
calibration intervals, CI-method comparisons, trend-power curves,
equivalence-power curves, covariance-assumption comparators,
decisive-classification safety metrics, checkpoint/resume support, and
validation figures.
- Added a focused BCa validation subset because BCa requires
leave-one-subject-out acceleration and is substantially more
computationally expensive.
- Added unit tests ensuring robust intervals are ordered/finite and
that changing the displayed CI method does not change analytic p-values
or omnibus inference.
- Version 0.2.0 is the confirmatory-validation build motivated by the
completed v0.1.9 benchmark, which showed strong core calibration but
mild heavy-tail undercoverage for analytic layer CIs.
OmicsBraid 0.1.9
- Added
test_braid_trend(), a covariance-aware GLS
trajectory test with a prespecified practical slope margin.
- Replaced raw observed-slope attenuation/amplification rules with
inferential trajectory states.
- Added hierarchical braid status: confirmed subtype,
direction-confirmed broader concordance, no-evidence, unresolved, and
insufficient.
- Added
no_detectable_effect to distinguish failure to
reject the joint null from demonstrated practical equivalence
(null_equivalent).
- Buffering/emergence remain confirmatory only when the required
layers pass equivalence testing; a separate
suggestive_pattern reports effect geometry when precision
is insufficient.
- Expanded simulation validation with Monte-Carlo intervals, trend
operating characteristics, exact versus hierarchical-family accuracy,
null-compatible outcomes, independence-assumption comparators,
scenario-failure reporting, and stratification by sample
size/correlation/missingness/distribution.
OmicsBraid 0.1.0
- Initial research implementation.