[t0mdavid-m]

This PR is stacked on #156. It needs OpenMS/streamlit-template#584, which adds the standard Viewer page (src/layout_manager/viewer.py) and the visualization library (src/viz/). This branch merges that template branch, so it runs on its own. It also merges main.

What changes

Standard Viewer page

content/viewer.py no longer builds the page out of the layout manager's pieces. It describes the runs as a ViewerSpec with one DatasetKind and calls show_viewer.

The template then draws:

• the experiment selector with its buttons;
• Single run | Comparison for an experiment with added runs: the whole experiment, or one run of it;
• the view bar with its views, pins and missing-analysis badges;
• every experiment side by side, each with its own Insight state.

Edit comparison opens the template's dialog (src/layout_manager/comparison.py), one tab per step: groups, filter, missing values, normalize, test.

Example workspace with a finished run

A new workspace starts with a processed example run, so the Viewer shows results right away without running the workflow first.

• example-data/workspaces/default/topp-workflow/ holds the run of the example files (results, logs, settings and experimental design). Its paths point at example-data/mzML/ and example-data/db/.
• settings.json names it as demo_workspaces.new_workspace. Every online workspace and the local default workspace start from it. Large files are linked, not copied.

Quantification	Identification

Everything computed during the run

The Viewer no longer processes anything on its first visit. At the end of a run, WorkflowTest.prepare_viewer() (src/viewer/precompute.py) fills <run>/viewer-cache/ with what the Viewer reads:

• the per-file overview and QC tables;
• the per-sample counts;
• the protein tables (proteins, peptides, PSMs) and the mass shifts;
• the abundance table;
• the group comparison with the recommended settings.

Each entry is keyed on the file it was computed from (size and modification time), so a rewritten file is computed again (src/common/disk_cache.py). A failure there never stops a run: the Viewer computes what is missing. Switching between the example views (Identification, Quality control, Pathways, Quantification) shows no spinner and no plot left loading.

Plots from the template's visualization library

The plots FLASHApp draws the same way now come from the template's src/viz/, with OpenDDA's own tables as their source:

• GO enrichment (Pathways view, go_panel.py): category selectors, the number of selected proteins on a line of its own with Selection | All proteins, and the enriched terms and the GO terms of all proteins as paginated Insight tables.
• Comparison steps in Edit comparison (comparison_steps.py): where values are missing, the values imputation fills in, the boxes per sample before and after normalization, and the p-value histogram. src/common/postprocessing_plots.py keeps only the filter curve, which only OpenDDA draws.
• Volcano, PCA and clustered heatmap (volcano.py, pca.py, heatmap.py).
• Group colours come from the library, so a group has the same colour in PCA, the clustered heatmap and FLASHApp's comparison plots.

The abundance profile, the plain heatmap and the comparison table stay in OpenDDA.

Layout manager

The "Earlier components" (run summary, identification quality, protein explorer) and "Files", which is not a visualization, are no longer offered. Their content is in the current components.

Cleaner viewer

• Titles and settings inside the components. The titles of PCA, heatmaps and the other plots are drawn by the components themselves. Their settings are in the component toolbar, for example the colorbar and the axes of a heatmap. Before, both were Streamlit widgets around the component.
• Captions and info boxes removed, such as "Counts at 1 % PSM FDR." on the Quality control view, and the GO enrichment info box in the middle of the plot.
• Info icons. Each component shows a ⓘ at its top left, with what it shows and how to use it. The texts are in src/viewer/teaching.py.
• GO terms computed with the run. The GO annotations are fetched when the run is processed, so the GO enrichment plot no longer loads on first view.
• Heatmap labels. Sample and protein names are shortened in the middle, and only as many as fit are labelled; the axis titles no longer overlap the tick labels.
• Fix: the heatmap's colour bounds are read as numbers. Before, they could fail with polars.exceptions.ComputeError: cannot compare string with numeric type (f64).
• Volcano. It opens on a symmetric range so the volcano is visible at first glance, and its selection buttons are in the plot toolbar. It ends level with the abundance plot beside it.
• Abundance per sample. The samples are ordered group by group, each group in its own colour, with a line at the group mean, and the plot is taller.

Depends on

• Four layers for code, a standard Viewer page apps describe as data, Single run | Comparison and Edit comparison streamlit-template#584.
• Help tooltips, heatmap settings and text axes, aligned volcano plot t0mdavid-m/OpenMS-Insight#107 (among others, SeriesPlot box and bar modes, used by the comparison steps). This branch uses its development build from dev-wheels/. Before a release, that must become a PyPI pin.

Screenshots (example workspace, light)

Pathways: GO panel, selection	GO panel, all proteins	Identification

Quantification	Quality control	Pathways (before this round)

Volcano and abundance per sample:

Before	After

The screenshots were added in one commit and removed in the next, so they are not part of the diff. Screenshots of the library's group colours in the clustered heatmap are still to come.

Tests

• Full suite (test_gui.py tests/) passes (1328), including tests/test_disk_cache.py, tests/test_go_panel.py, tests/test_viewer_plots.py (now on the library's data functions) and the abundance profile test in tests/test_viewer.py.
• pylint (errors only) is clean on the changed files.

🤖 Generated with Claude Code

https://claude.ai/code/session_01FkEGen2HSCa8BFE1JW4jxN

[t0mdavid-m]

The desktop build's Windows end-to-end test failed in the rescoring scenario. The app process crashed with a native access violation inside polars.scan_parquet, while openms_insight was loading the protein abundance table's cache (src/viewer/panels/abundance.py:60), and the Viewer then timed out after 600 s.

The same scenario passed on Windows one commit earlier (run 43622561636, 2af3933). The commit in between only changes the QC layout and does not touch that table or how its cache is read, so this looks intermittent rather than caused by this PR. I'll rerun the failed job once the macOS job of this run has finished. If it fails again, it needs a closer look in openms_insight/core/storage.py.

---

Generated by Claude Code