⚠️ SERVICE NOTICE: This instance will be decommissioned 1 december of this year (2026).
Please migrate your projects to our new repository:
https://gitrepo.service.rug.nl
Sites now go from systemic, IV, meningitis, endocarditis, pneumonia and mastitis to UTI, topical and screen. A breakpoint without a site takes the position of "other", as both mean any site that is not listed next to it. A second site breaks a tie on the first one.
The veterinary sites are now fixed codes as well: mastitis (includes Mammary gland), respiratory, metritis, genital, soft tissue and wounds (includes Abscesses). Sites that are part of each other do not contradict.
site = "other" now leaves out every breakpoint limited to an indication, and prefers a breakpoint without a site.
A rebuild of clinical_breakpoints with this order gives the same data, so the same blocking rows stay.
The rebuild uses the WHONET files of today, and holds the changes of #326, #327, #328, #329 and #331. The files in data-raw/datasets are not regenerated yet.
The new internal blocking_row_is_needed() defines this: as.sir() must give another result without the row, so the definition follows the fallback path of as.sir() itself. The build removes all rows without an effect, and the new strict test checks all rows of the latest guideline, and for all years that a blocking row has a breakpoint to block.
This replaces the separate taxonomic comparison, handles the points of the review on this pruning, and keeps the sites of blocking rows, which `site` in as.sir() now uses.
Added: the oxacillin disk for S. pseudintermedius and related species, the oxacillin MIC for coagulase-negative staphylococci, mupirocin for nasal decolonisation, and meropenem in other infections for N. meningitidis. The Bacteroides statement is now matched as a pattern, and other wording stops the script.
The script documents why statements are left out, e.g., the oxacillin statements for S. aureus, S. lugdunensis and S. saprophyticus are hedged ("mostly methicillin resistant"), unlike those for cefoxitin.
Implements the decisions ticked in #328, and the behaviour specified in https://github.com/msberends/AMR/pull/339#issuecomment-6577438074: `include_screening` now defaults to NULL, using screening breakpoints for screening agents only. With TRUE, a regular agent also takes these breakpoints, and with FALSE a screening agent gives NA with a note.
- antimicrobials gains 13 screening agents, e.g. GEN-S, ERY-S and STR1-S
- GEH and STH rows move to GEN-S and STR1-S, and as.sir() keeps accepting both codes
- Topical cut-off values move to -S agents with site "Topical"
- WHONET-based rows move as ticked, including CLSI cefoxitin for staphylococci and pefloxacin for Salmonella, and EUCAST 2012-2018 cefoxitin MIC
- Clinical breakpoints that EUCAST also allows as a screening test get a copy under the -S agent
as.sir() treats them as absent, so it can use PK/PD breakpoints or, for unknown `uti`, UTI breakpoints. The values stay available in data-raw/breakpoints_eucast_raw.csv, to which the manual now points. Organisms that a table excludes stay excluded, since these have their own blocking rows.
This also fixes the change for unknown `uti` of the previous commit: a blocking row only gives way to an actual UTI breakpoint, not to a UTI row without breakpoints.
`site` takes fixed codes, e.g. "meningitis" or "iv", and can be a column. With `site = NA`, any fitting breakpoint can apply, from most to least systemic. With a site set, breakpoints for contradicting sites are left out, while a breakpoint without a site still applies, and the result is NA if only contradicting sites exist, without falling back to PK/PD. `uti = TRUE` stays available and equals `site = "uti"`. Every choice is noted in sir_interpretation_history(), including the site that was used.
For unknown `uti`, a blocking row without a UTI status no longer overrides a UTI breakpoint at the same taxonomic level. A `uti` vector mixing NA with TRUE or FALSE no longer gives an error.
Veterinary sites are not part of the fixed codes yet, and are treated as breakpoints without a site.
The reproduction script no longer maps MIC breakpoints to AMR:::COMMON_MIC_VALUES, and also keeps the off-scale S breakpoint of 0.001 mg/L. as.sir() compares through the new internal mic_dilution_level(), so that e.g. an MIC of 0.032 and a breakpoint of 0.03 are the same dilution, and shows the published breakpoints in sir_interpretation_history().
Every parsed breakpoint now holds the address of its workbook cell in breakpoints_eucast_raw, and the script stops if a second reading of the workbooks with readxl does not find the value in that cell.
The data sets are rebuilt once, at the end of this PR.
No code change was needed. Morganella follows Enterobacterales, except for imipenem and imipenem-relebactam, for which EUCAST gives rows for Morganellaceae or Morganella.
The new compare_clinical_breakpoints.R compares the interpretations of two builds, and fails on any change that is not explained. A rebuild of main with pinned WHONET files reproduces all rows of clinical_breakpoints and microorganisms.codes.
The filter on MIC rows in the scope extensions is removed, since these now join on the method of each statement. It kept the DISK rows of the Bacteroides statement from being copied.
Plotting:
- Base colours and legends of interpreted MIC and disk values on the interpretation itself instead of matching colours
- Fix SDD in autoplot() for disks, the colours of barplot() for SIR values, and barplot() for MICs ignoring language and expand
- Match a named colours_SIR by name; let keep_operators = TRUE/FALSE in scale_*_mic() behave as "all"/"none"
- Let a named values in scale_*_sir() replace the colours it names, warn in scale_x_sir(), and pass other arguments on
- Let the default of eucast_I follow the guideline option, also for EUCAST before 2019
- Support unquoted column names in facet_sir() without evaluating them
- Correct the documentation and examples
Translations:
- Add SDD, NS, WT and NWT, and correct S, I, R and NI in 18 languages
- Sources of the terms for 'I': NAK via German laboratories (de), SZU national reference laboratory (cs), lex.dk (da), Karolinska University Hospital (sv), COESANT (es), Nordlab (fi), CA-SFM (fr); no by analogy with da and sv
- Not verified against an official source: ar, bn, el, hi, id, ja, ko, pl, ro, ru, sw, tr, uk, ur, vi, zh
Tests:
- Let the grouped antibiogram test use English so it passes in other locales
WHONET names are now matched exactly (also outdated names, after removing qualifiers), then on GBIF identifier, and fuzzy results are only kept if consistent with that; otherwise the genus or UNKNOWN is used. A code refers to the record of the name it was named after (e.g. EAE is Enterobacter aerogenes) if that name refers to the same taxon. Every remap of an existing code to another taxon must be listed in data-raw/whonet_remaps_reviewed.csv; the 198 remaps of this rebuild were reviewed.
The full rebuild of clinical_breakpoints also assigns Salmonella serovars and the Bacteroides fragilis complex correctly and adds ECOFFs for Neocosmospora solani. The rebuilt microorganisms.codes is not included, since it remaps WHONET codes that need review first.
The pruning of blocking rows in the EUCAST reproduction script now compares UTI status too, and the script is fixed for the renamed column of the as.ab() coercion cache. In clinical_breakpoints, only the blocking rows were replaced (356 organism/agent/method combinations newly blocked, none unblocked); other changes of a full rebuild, caused by taxonomy updates in the WHONET-based rows, are not included.
- New sir_combine() to combine SIR results of multiple antimicrobials into one variable; antibiogram(), wisca() and the count/proportion functions now use the same engine (antibiogram() and wisca() output unchanged)
- With multiple antimicrobials, count_*() and proportion_*() now follow the documented combination therapy rules: resistance() is the proportion of failing combinations, count_all() excludes isolates with an unknown combination result, and WT/NWT/NS/NI are handled correctly
- where() is now decided at runtime instead of at installation, fixing antibiogram() and wisca() without tidyselect (e.g. CRAN binaries without dplyr)
- Imports from data.table and cli moved to .onLoad(); data.table functions are only used while it is loaded; the like() example no longer unloads data.table, which crashed R
- Static test that fails on code outside functions depending on suggested packages
- New check-without-dplyr workflow: installs AMR with all suggested packages, removes dplyr (and tidyselect), then runs all unit tests and examples against the installed package
- g.test() now returns the degrees of freedom as 'parameter', so they are printed
- Fix R CMD check NOTEs (hidden .claude folder, lost braces in inherited g.test and ggplot_pca docs)
Package:
- stop_() called cli::ansi_has_hyperlink_support() before checking that cli is available, so without cli every error became "there is no package called 'cli'"
- interpretive_rules() failed on R 3.0 for guidelines without AmpC rules (assignment to zero rows)
- Guard the ggplot2 example of wisca_plot()
CI:
- check-old: R 3.0 could not install tinytest over HTTPS, so no unit tests ran; install it with curl, and fail when tinytest is missing or any test fails
- tinytest shim: record expectations in the test file environment, support skips and errors per test_that() block, add missing testthat equivalents, source helper files
- Add R 4.0 (oldrel-6) to check-old, and oldrel-5 to check-recent
The pre-commit URL extractor captured the closing brace of \url{}, which
made the EC server redirect to the dead health.ec.europa.eu domain, after
which urlchecker::url_update() wrote that dead URL back into R/ab.R and R/av.R.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Aerococcus rules now use `one_of`, so multi-species rules (breakpoints v12 onwards) apply
- Pasteurella rules filed under Neisseria meningitidis moved to their own rule
- Pasteurella rules now apply on genus level
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- test-data-microorganisms.R: warnings are wrapped to the console width, so the expected patterns now
allow a line break between words (failed on GitHub Actions since v3.0.1.9097)
- reproduction_of_microorganisms_CLAUDE.md, step 4: the agent never changes the version number or date,
the human sets them when merging
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- the list of known defects is development data and must not ship with the package
- tests read it from the source repository; mo_repository_root() now walks up from the working
directory, so the tests also find the repository from an R CMD check directory inside it (GitHub Actions)
- the integrity, known-defects and sentinel tests skip when the source repository is not available
- CLAUDE.md: on local commits to main, the git hooks set the version number and date
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
As curated in data-raw/Species groups.xlsx since July 2023; the released v3.0.1 data still
used 'Mycobacterium avium-intracellulare complex' for the same group of organisms.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- taxonomy build: released taxa (since v2.0.0) that are missing are restored from their
last release with their old code and last known status, and are exempt from later
removals (decision by Matthijs S. Berends)
- exceptions are retired on purpose in data-raw/microorganisms_files/mo_code_retirements.csv
with a reason: other organisms with the same genus name (e.g. the Graphium butterflies,
wrongly in the Fungi until v3.0.1) and taxa whose genus is now only known in another
domain; as.mo() translates these codes to NA with that reason
- synonyms without a current name are linked to an accepted name in the same genus that
only differs by a Latin gender ending (e.g. Candida glabratus -> Nakaseomyces glabratus)
- placeholders such as "(unknown class)" never become records; codes of manually added
records are matched by domain and name
- integrity rule released_taxon_missing; known defects of the data on main updated
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- data-raw/microorganisms_files/: registry of all MO codes of every release since v2.0.0
(78,992 codes, built from the git tags), approved renames, and README with the rules;
decision by Matthijs S. Berends: MO codes from before v2.0.0 are not supported
- as.mo(): codes of earlier releases are translated deterministically via the internal
MO_RETIRED_CODES (generated from the registry); unknown codes are no longer guessed
(previously F_TRCHP, Trichophyton in v3.0.1, became Talaromyces chlamydosporus)
- synonym_mo_to_accepted_mo(): a missing target of one source no longer overwrites a valid
target of another source (e.g. Chaetomium abuense), and no 'renamed to itself'
- taxonomy build: codes are reused and reserved from the registry instead of the development
data (fixes 40 codes reused in the 2026 development data), also for phylum to family;
synonym cycles are broken; dangling 'renamed to' identifiers are removed; the build stops
on any broken integrity rule before saving
- run_microorganisms_build.R: runs the build in chunks at its checkpoints in one R process,
with progress logging and resuming from any checkpoint; prompt updated accordingly
- tests: shared integrity rules (helper-microorganisms.R) used by build and testthat, registry
tests over all codes, sentinel organisms, regression tests; current defects of the data on
main are listed in microorganisms_known_defects.csv until the next rebuild
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Review of reproduction_of_microorganisms.R against the 2026 intermediates:
- genus homonyms: overrides, LPSN/MycoBank first, status within LPSN/MycoBank,
move misplaced protists, remove true homonyms (Necator, Capillaria, Graphium)
- keep clinically relevant names that are not validly published (Tropheryma)
- one set of relevant genera for all filters; 30 genera added to MO_RELEVANT_GENERA
- MycoBank classification parsed from all columns; empty values never overwrite
- GBIF synonyms in another genus get the lineage of their own genus
- prevalence no longer copied from genus; current names inherit from synonyms
- inferred records follow the relevance filters (no more unrelated lichens)
- ASCII names before deduplication (12 duplicate full names in 2026)
- MO codes: robust code parsing, reserved old codes, single allocator,
domain column used directly (2026 rename would now fail)
- name clean-up applied to taxonomy before MO codes, BacDive rows filled
- unattended running: review() instead of View(), CSV export of reviews,
LPSN cache and retries, stops when raw source files are missing
- reproduction_of_microorganisms_CLAUDE.md: prompt for an unattended run with
findings report and checklist for human review in the PR
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Genus resolution now prefers LPSN/MycoBank over GBIF, then accepted over
synonym, then source, then rank, and only then domain. Losing LPSN/MycoBank
homonyms are removed instead of moved into another domain. Adds integrity
checks that LPSN records stay prokaryotic and MycoBank records stay fungal.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>