[amjjbonvin]

What does this PR do and why?

Adds an HTML report, so the CSB portal can show DisVis results with a Mol* view, as it does for PowerFit, and drop the Chimera images (#46).

DisVis now writes report.html and state.mvsj (MolViewSpec, Mol* 5.13.1 mvs-stories from jsDelivr, as in powerfit_em/report.py) in the output directory, and copies the input files there (the report shows the fixed chain and links all of them); --no-report turns this off. disvis.log and the terminal end with how to view the report. The report shows what the portal's results page shows now:

• accessible complexes per number of consistent restraints (N);
• violations per restraint, without the all-zero rows;
• z-scores, labelled with the restraints (e.g. A27(CA)-A18(CA)), z ≥ 1 highlighted;
• with -is, for each N the 10 most contacted receptor and ligand residues (most first), and the full interaction tables, colored from white to green (most contacted);
• a Mol* view of the fixed chain with accessible_interaction_space.mrc for each N and, with -oa, occupancy_N.mrc (green); with -is, the selected residues of the fixed chain are colored from white to red by their average number of interactions per complex.

The tables (except the lists of most contacted residues) sort by a column when its header is clicked (▲/▼ on the sorted column). All files are referred to by relative names. The report is built from the output files, which are unchanged. New dependency: molviewspec.

disvis.log now starts with a header giving the DisVis version.

Each example-results directory has its report (generated from the existing reference files). New reference run PRE5-PUP2-complex/example-results/cpu/coarse-is-oa (-is selection.res -oa), to show the occupancy maps and interaction analysis; its other output files are identical to cpu/coarse.

How was this tested?

• New unit test (tests/test_report.py): generate_report on the text output of a small -is -oa run (tests/data/output); checks one snapshot per N, relative URLs only, the restraint labels and interaction tables. All 27 unit tests pass (-W error), also in the Pyodide wheel (local cibuildwheel).
• test-cases/check_test_cases.py (coarse, now with the report): output files byte-identical to the references.
• Reports checked in headless Chrome over HTTP (PRE5-PUP2 coarse-is-oa, RNA polymerase II fine). The residue colors are not checked by the unit test.

Not tested: the portal embedding. Opened from disk (file://), the browser blocks the files of the viewer; the page then says to serve the directory with python -m http.server.

AI assistance

Written with Claude Code; checked as above.

Checklist

• Tests cover the new and/or changed code
• Documentation updated if needed (README.md, INSTALLATION.md)
• Reference results in test-cases/*/example-results/ updated if the output changes (output unchanged; reports and the coarse-is-oa run added)

Related issues

Closes #46

Notes for reviewers

• The issue mentions accessible_interaction_space_<N>.mrc; DisVis writes a single map whose value is the number of consistent restraints, so the report shows it at level N − 0.5 for each N (Chimera in the portal uses level N).
• Occupancy maps are shown at 2 sigma, in green over the grey interaction space; the view is centered on the fixed chain (Mol* would otherwise fit the whole map grid). Colours and levels are open for suggestions.
• molviewspec 1.8 raises a pydantic deprecation warning on import, hidden in normal runs; the test ignores it (the unit tests run with -W error).

🤖 Generated with Claude Code

[sverhoeven]

Works for me.
I was able to generate the report and use --no-report

Regarding inline comments.
There are 3 requests for changes: README.md, slow test and http server log.
The other comments are suggestions.

Btw congrats on https://pypi.org/project/disvis/