Skip to contents

Changelog

Source: NEWS.md

CytokineProfileShinyApp 0.0.1

To be updated.

CytokineProfileShinyApp 0.0.0.9000

This development build focuses on making the app easier to use, easier to read, and more reliable.

What’s New

  • You can now run Two-way ANOVA and ANCOVA directly in the app.
  • ANCOVA now supports optional secondary:covariate interactions, with clearer slope checks and more appropriate follow-up comparisons when factor-by-covariate effects are included.
  • The app now gives clearer warnings when interaction effects may make results harder to interpret on their own.
  • Many figure-based analyses now include easier text controls, so you can adjust titles, axis labels, legends, and other plot text without digging into code.
  • Random Forest and XGBoost outputs now report model setup details more clearly, including split settings, class balance, cross-validation status, and key tuning choices.

Easier to Use

  • Re-uploading a file now refreshes the saved copy right away, even if the file has the same name as before.
  • Excel imports are more reliable. If a previously chosen sheet is no longer available, the app now switches safely instead of failing.
  • When you upload a new file, saved editor choices from the previous file are cleared more cleanly.
  • Columns that look like numbers, even when they include symbols or common missing-value markers, are now more likely to be recognized correctly.
  • Step 2 now lets you force selected columns back to numeric, and any leftover non-numeric entries are converted to missing values instead of blocking numeric workflows.
  • Column selections now update more smoothly, and related defaults stay in sync better as you change your data choices.
  • The missing-value help now gives clearer guidance on when to use mean, median, mode, and the two nearest-neighbor options, and the popup is easier to read.
  • Two-group testing controls now label Welch t-test more explicitly, so the manual choice matches the app’s underlying behavior more clearly.
  • PLSR now handles partially missing predictor data more gracefully by keeping usable rows, dropping only unusable predictors, and giving clearer guidance when sparse columns may need missing-value treatment first.
  • sPLS-DA now handles partially missing predictor data more gracefully by dropping unusable predictors, keeping rows that still contain retained predictors, and warning more clearly when sparse columns may need missing-value treatment first.
  • If your uploaded data includes out-of-range values, the app now warns you when you click Save & Use and explains what was adjusted in plain language.
  • The app now starts more reliably across repo-root, installed-app, and package launcher workflows.
  • Step 2 now restores its state more consistently when you move forward and then return, including dynamic categorical filter selections.
  • Fresh-start, reuse-data, upload, built-in editor, and Bio-Plex workflows now behave more consistently after recovery and cleanup work.
  • Progress notifications are easier to read during longer analyses, with cleaner spacing and clearer separation between the main task and detail text.

Better Figures

  • Plot text is now more customizable across many figure types.
  • PCA, PLSR, sPLS-DA, and MINT sPLS-DA figures now use more readable default text sizes.
  • Correlation circle plots in PCA and MINT sPLS-DA now behave more reliably.
  • Heatmaps now display more consistently during interactive use.
  • Volcano plot now displays p-value in different sizes depending on how small the p-values are. Additionally, it shows upregulated vs. downregulated variables clearly.

Bug Fixes

  • Some result tables are now labeled more accurately when column names contain underscores.
  • The app handles empty analysis results more gracefully instead of returning confusing output.
  • The XGBoost workflow now chooses the best training step more reliably across different scoring methods.
  • Editing data in the built-in editor now preserves values more consistently.
  • The missing-value help popup now opens more cleanly and is less likely to be cut off inside the app window.
  • The missing-value method selector now keeps its nearest-neighbor options in sync more reliably instead of dropping or mismatching those controls.
  • PLSR is now more stable when only one component can be fit, and the VIP>1 follow-up preview now skips safely instead of failing when too few predictors remain.
  • sPLS-DA is now more stable when Excel uploads contain empty cells, and the VIP>1 follow-up preview now skips safely instead of failing when too few predictors remain above the threshold.
  • Deselecting all categorical columns in Step 2 now stays deselected instead of snapping back to all selected.
  • Step 2 dynamic categorical filter selections now restore correctly after you return from Step 3.
  • kNN (feature-wise) now safely blocks unsupported single-column use with a controlled message instead of a raw failure.
  • Running an analysis from Step 4 no longer jumps to Step 5 before the analysis actually succeeds, so failed runs stay on the inputs screen with a clearer error instead of looking completed.
  • Input validation and export failures now use friendlier app messages, and common missing analysis settings are checked before model code runs.

Behind the Scenes

  • Several parts of the app were cleaned up to reduce small startup and package-loading problems.
  • Internal helper code and UI regression checks were cleaned up so package checks behave more consistently across local testing and staged installs.
  • The server pipeline was re-extracted into dedicated mod_*_server() stage files, with shared stage helpers now centralized in R/app_stage_helpers.R.