dar 2.0.0
Changes since dar 1.8.0.
Breaking changes
- R 4.6 or later is required. Differential-abundance recipes configure their targets and taxonomic rank with
add_model(). The old selectors and recipe slots are removed. Rebuild saved recipes from their inputs. - Use
step_aldex2()for ALDEx2 andrecipe_qc()for quality control.step_aldex(),phy_qc()andstep_metagenomeseq()are removed. ALDEx2 effects now use the fitted GLM coefficient. - Recipes use a versioned JSON exchange format. Importing bakes into an unprepared recipe requires
prepare = TRUE. -
otu_table()returns counts with taxa in rows. Sample and taxonomy accessors return the full tables. ComplexHeatmap, gplots, heatmaply and UpSetR are optional.
Models and results
- Shared models support covariate adjustment, longitudinal contrasts and compatible mixed designs. Added LinDA, ALDEx3 and radEmu, with per-stage
engine_argsfor advanced options. -
tidy_results(),tidy_abundance(),get_execution()andeffect_plt()expose per-method results, aggregated abundances, execution details and effect estimates. - Abundance plots support shared or union discoveries and explicit taxa. Automatic selection requires a contrast when ambiguous. Consensus outputs retain the contrast and direction of each discovery.
Fixes and maintenance
- Random-intercept models accept singleton participants alongside participants with repeated observations. Checks use observed groups and still reject designs with one group or no repeated observations (#189).
- Recipes validate steps before execution, run preprocessing in declared order, reserve result column names and assign unique automatic IDs (#187).
- DESeq2 uses native
poscountsnormalization and applies the significance threshold and result options before shrinkage (#188). - Corrected contrast handling in MaAsLin3 and LEfSe, robust inference in ANCOM-BC2 and corncob, and depth-dependent zero handling in LinDA.
- Five vignettes document the current workflow. Docker Hub publication runs separately from GitHub package checks (#190).
dar 1.5.5
Breaking changes
-
step_maaslin()uses MaAsLin3 in place of MaAsLin2. Review method-specific arguments when migrating existing recipes.
dar 1.5.4
- Parallel execution limits connection setup to support servers with more than 128 cores.
- Parallel execution uses a temporary future plan, preserving the user’s global future configuration.
dar 1.1.3
-
step_ancom()is temporarily disabled.
dar 0.99.10
-
step_corncob()is reintroduced after corncob returns to CRAN.
dar 0.99.9
-
prep()uses four workers to avoid issues with Bioconductor build infrastructure.
dar 0.99.6
- Dependency installation guidance recommends
BiocManager::install().
dar 0.99.4
- Tests and examples take less time to execute.
- The Linux GitHub Actions workflow correctly installs rlang.
dar 0.99.3
Breaking changes
-
RecipeandPrepRecipereplace the formerrecipeandprep_recipeclasses. Update code that inspects the old class names. -
required_deps()is internal and is no longer exported. Follow the installation guidance emitted by recipe steps for missing dependencies. - R 4.4.0 or later is required.
- The package stops re-exporting
%>%and:=. Use|>in recipes and qualify functions from their originating packages when needed.
New features
-
recipe()accepts both phyloseq and TreeSummarizedExperiment objects. - Filtering and subsetting steps support chainable recipe construction.
-
step_filter_by_abundance(),step_filter_by_prevalence(),step_filter_by_rarity()andstep_filter_by_variance()provide dedicated filtering criteria.
Documentation and improvements
- The import tutorial covers BIOM, QIIME, mothur, MetaPhlAn and HUMAnN conversions to phyloseq and TreeSummarizedExperiment.
- Installation guidance uses
BiocManager::install()andinstall.packages()instead of unconventional installation methods. - The data.table dependency is removed.
- Rarefaction warns about its fixed seed and the implications for results.
