[joshua-d-campbell]

What changed and why

The maintainer asked for these Shiny app fixes. AGENTS.md otherwise keeps
app changes propose-only.

Celda tab feature selection. Both methods failed:

• Scran_modelGeneVar called scranModelGeneVar(), which doesn't exist.
  It now calls runModelGeneVar(). It also normalized with the deprecated
  scater::logNormCounts(), and now uses scaterlogNormCounts().
• SeuratFindHVG never ran. The server checked for "runSeuratFindHVG",
  but the dropdown's value is "SeuratFindHVG".
• Both passed n = to getTopHVG(), which takes hvgNumber. They now also
  pass useFeatureSubset = NULL, so the genes come from the computed
  metrics, and featureDisplay = NULL, so they come back as rownames that
  can subset the object.
• Both replaced the app's whole dataset with the selected genes, so every
  other tab lost the rest (found by /code-review). The HVGs, filtered by
  the count thresholds, now go only into the featureSubset altExp that
  Celda uses, as with the "None" option.
• With Seurat 5, only the vst method of SeuratFindHVG works. The other
  two fail inside runSeuratFindHVG() (runSeuratFindHVG fails for dispersion and mean.var.plot with Seurat 5 #804), and NEWS says so.

Starting the app with data. singleCellTK(inSCE = sce) returned HTTP 500
for any object without alternative experiments: ui.R called
names(altExp(x)). It now calls altExpNames(x).

Other changes:

• The SoupX help page describes the scrapper/bluster clustering from Move modelGeneVar and SoupX clustering to scrapper (results change) #800.
• The Celda workflow article names the Seurat option as the app shows it.
• Version 2.23.4.

How it was tested

• Manual check (maintainer): ran the Celda workflow in the app with both
  feature-selection methods; it worked. This was before the change that
  keeps all genes in the main dataset, which was verified only by script
  (next point).
• Keep-all-genes change (scripted): both branches keep the full
  200 × 195 main object, with 95 or 100 HVGs in featureSubset, and
  recursiveSplitModule() runs on the result.
• Scripted: the exact server-side calls for both branches ran on
  scExample with featureDisplay set:
  • modelGeneVar: 95 HVGs, returned as rownames;
  • SeuratFindHVG: 100 HVGs, returned as rownames;
  • the old calls fail with "unused argument (n = n)" and "could not find
    function scranModelGeneVar".
• App startup with preloaded data: HTTP 500 before the altExpNames()
  fix, HTTP 200 after.
• Automated UI test: a shinytest2 run couldn't click through the app,
  because it never becomes idle in headless Chrome (Shiny app never becomes idle under shinytest2, blocking automated tests #803).
• lintr is clean on changed lines, and server.R parses.

Issues found while testing and left for later:

• Shiny app logs errors from unopened tabs when started with preloaded data #801, Shiny app: console auto-scroll script error and duplicate HTML ids #802, Shiny app never becomes idle under shinytest2, blocking automated tests #803: app startup, front-end, and test infrastructure.
• runSeuratFindHVG fails for dispersion and mean.var.plot with Seurat 5 #804: runSeuratFindHVG() fails with Seurat 5 for dispersion and
  mean.var.plot.
• Celda tab runs Seurat vst HVG selection on normalized data instead of raw counts #805: the Celda tab runs vst on normalized data. Tracked rather than
  changed, by maintainer decision.

Checklist

• Tests added or updated, make test passes, and make coverage
  didn't drop. Not applicable: only inst/shiny changed, and there is
  no app test suite yet (Shiny app never becomes idle under shinytest2, blocking automated tests #803).
• make check-full and make bioccheck pass with no new errors or
  warnings. Not run for this PR: only inst/shiny, NEWS, DESCRIPTION,
  and an article changed. Move modelGeneVar and SoupX clustering to scrapper (results change) #800's runs covered R/.
• make docs run, if roxygen comments changed. N/A.
• New exports added to _pkgdown.yml, and make site-check passes.
  N/A.
• NEWS.md updated for user-facing changes
• Version bumped (z) if this will be pushed to Bioconductor
• Plan review and /code-review run; findings fixed or answered.
  /code-review found three issues: one fixed here, two tracked in
  runSeuratFindHVG fails for dispersion and mean.var.plot with Seurat 5 #804 and Celda tab runs Seurat vst HVG selection on normalized data instead of raw counts #805.
• Related issue linked: Shiny app logs errors from unopened tabs when started with preloaded data #801–Celda tab runs Seurat vst HVG selection on normalized data instead of raw counts #805
• Shiny only: verified in the running app (maintainer's manual check)

ADR

N/A: no structural change.

Scientific correctness

Generated content

Claude Code (Opus 5.5) wrote all of this PR. The maintainer requested the
app changes and checked them in the running app.

🤖 Generated with Claude Code