NEWS

---

woven 0.99.3

• Removed a stale DESCRIPTION claim about sparse projection matrices (PMD); no such code path exists (woven() recovers dense projection matrices by design, optionally ridge-regularized and feature-screened via ridge_w and screen_top). DESCRIPTION now describes that mechanism instead.
• .scale_fit() (internal) rewritten without <<- (BiocCheck NOTE).
• summary()'s printed report now goes through a dedicated internal wovenSummary S4 class with its own show method, so cat() output only ever appears inside a show method (BiocCheck NOTE). summary() still returns the metrics vector invisibly; printed output is unchanged.
• Escaped literal braces in a solver_mcca_dual.R roxygen comment that R CMD check was parsing as Rd markup ("Lost braces" NOTE).
• .Rbuildignore now excludes *.Rcheck/*.BiocCheck directories, so a leftover local check-output directory can no longer ship inside the built source tarball (one had shipped in the 0.99.2 tarball).
• Added tests/testthat/test-precompute-plot-summary.R, covering woven_precompute() reuse/equivalence, a malformed-precomp error path, summary() with and without labels, plot() with/without labels and anchor-highlighting (including its out-of-range-dims error), and woven_nystrom_error() as a regression test for the 0.99.2 Za_list field-name bug fix. These close out the last open items from the agentic Bioconductor review's testing-gap finding.
• Verified clean from a rebuilt tarball: R CMD check 0 errors/0 warnings/0 notes; BiocCheck 0 errors/0 warnings/4 notes (down from 6, all cosmetic); 88/88 tests pass.

woven 0.99.2

• The object returned by woven() is now an S4 class (woven, defined in AllClasses.R) instead of a plain S3 list. show, summary, and plot are now real S4 methods. New accessor generics Z(), WList(), anchorIdx(), and singularValues() are the recommended way to pull the most commonly used slots out of a fit; see ?\woven-class`for the complete slot list, or use@` directly for anything an accessor does not cover.
• Fixed a real bug in woven_nystrom_error(): it referenced a field (fit$Za_list) that did not exist on the fit object (the correct name is Z_anchors), so the function always silently returned NaN rather than ever computing a real projection error. It also passed a too-short Y vector to the internal leave-one-out refit; both are fixed, and the function now returns a real, non-degenerate value.
• plot() on a woven fit now returns the ggplot object directly (visibly) instead of forcing a print() call internally and returning it invisibly; this is the standard R plot-method convention and avoids printing outside a show method.
• woven() now accepts a MultiAssayExperiment directly as X_list. Each experiment becomes one modality; subjects are aligned to colData via sampleMap, and a subject absent from a given experiment gets an all-NA row for it automatically, matching the existing block-missing convention. Y may be a colData column name (a single string) instead of an explicit vector. See ?woven for a worked example and .mae_to_xlist() for the conversion internals.
• Laplacian construction in woven_precompute() now parallelizes via BiocParallel (MulticoreParam/SnowParam/SerialParam chosen by platform) instead of parallel::mclapply, which does not fork on Windows.
• Fixed a stale ?woven description of solver routing (it described a since-removed V=2-only closed-form / V>=3 ALS split; the package has used a single closed-form dual SUMCOR MCCA solver for all V since 0.99.0).
• Fixed "Nystrm" typo (-> "Nystrom") in source comments and documentation.
• Vignette introduction rewritten: defines "assay block", explains the Nystrom method and its use as an out-of-sample extension (with a link to further reading), and states WOVEN's comparison to DIABLO concretely instead of via an unqualified "standard methods" claim.
• (Retroactive note, first available in 0.99.1) woven() exposes ridge_w and screen_top, ridge-regularized / feature-screened recovery of the projection matrices, useful when a modality has many more features than anchor subjects and the true signal is diffuse. Select jointly via cross-validation on held-out classification error.

woven 0.99.1

• woven() now standardizes each modality to zero mean and unit variance by default (scale = TRUE), matching the internal scaling of comparator methods so that no modality dominates by measurement scale. Stored centers and scales are reapplied to new subjects in woven_scores() and woven_predict(). Set scale = FALSE to retain the previous raw-feature behavior.

woven 0.99.0

• Initial Bioconductor submission.
• Unified dual SUMCOR MCCA solver (woven_mcca_dual) for all V >= 2 modalities.
• Nyström out-of-sample extension for block-missing subjects.
• Graph Laplacian regularization built on observed rows only (no imputation).
• Label-augmented cross-covariance for supervised integration.
• Full benchmark against DIABLO, MOFA2, IntegrAO, and Impute+DIABLO across four simulation arms and ADNI real-data validation.