Simplify scoring, report and biophysics code without changing outputs

[DimaMolod]

A simplification pass over the package: dead code, duplicated logic and needless indirection. No output changes. Net −1,257 lines (920 added, 2,177 removed). report.py goes from 2,031 to about 1,370 lines and connolly.py from 910 to 750.

This targets the #30 branch, so the diff shows only the cleanup.

Verification

• Golden outputs. The CLI ran over every bundled run (AF2 pos/neg, AF3 pos/neg, AF3x 4G3Y crosslink fixture, Boltz-2 6OGE). The runs covered all models, per-run and aggregate PDFs, cached-CSV reuse, --skip_biophysical_scores, non-default thresholds, and the alphajudge-report entry point. Report timestamps were frozen for the comparison. All 63 CSV, PDF and PNG files are byte-identical to the base commit. The base itself was checked to be deterministic across two runs.
• Biophysics. Raw Connolly dots, normals, burial flags and molecule labels, plus SC, H-bond, salt-bridge, disulfide, buried-area and solvation values, are bitwise equal on 16 interfaces.
• Tests. pytest test/ passes 92/92 with ALPHAJUDGE_RUN_SLOW_SC_REFERENCE=1.

What changed

• report.py. Removed:

  • the page-total and footer plumbing that nothing read;
  • the never-drawn logo, info icon and slider-marker branch;
  • an unused row grouper and three unused imports.

  Repeated axes setup and text boilerplate now go through _text_axes, _label_axes, _kv_rows and _save_page. The aggregate report no longer sorts lists that are already sorted, and counts backends with Counter.

• runner.py.

  • One functools.partial worker serves both the serial and process-pool paths.
  • The cache-reuse and recompute branches share the report and aggregation tail.
  • Rows are built in _interface_row.
  • Removed the unused matplotlib/numpy imports and a lazy import of a module that was already imported.

• Interface / Complex.

  • _pair_indices() replaces four copies of the residue-to-PAE lookup.
  • mpDockQ's global contact count is now the sum of the per-interface counts. Biopython's KD-tree applies the same d² <= r² test, so a second neighbour search is gone.
  • Removed a per-chain-pair atom cache that could never hit.
  • New Interface.label and Complex.chain_boundaries replace private-attribute access from the runner.

• Parsers.

  • JSON and pickle loading share magic-byte decompression.
  • Path lookups share BaseParser._first_existing.
  • ParserManager keeps only register and pick.

• Biophysics.

  • bonds: one KD-tree contact generator for H-bonds, salt bridges and disulfides.
  • prosurf: one coverage helper instead of two copied loops.
  • connolly: removed the unused trim(), _disptl and a neighbour list that was never read; shared probe-torus geometry between its two users.
  • sc, solvation: less bookkeeping.

• scripts/. Smaller validate_ccp4_biophysics.py parsing, and the PISA session is now erased even if -analyse fails. STANDARD_AA is a single line.

Removed API, all with no callers (noted in CHANGELOG):

• ParserManager.unregister, enable_only, set_precedence, list_parsers
• confident_contacts.representative_atom_contact_pairs
• biophysics.connolly.trim
• the ignored residue_names argument of connolly.mds

alphajudge.core and alphajudge.biophysics.scoring still import as before.

Follow-up

The problems found during this review are fixed in the stacked draft PR #32. They change results, so they are kept out of this no-output-change PR:

• solvation energy now matches PISA;
• Boltz-2 PAE, pLDDT and pair ipTM are aligned to residues by token and chain;
• report fixes for the PAE lookup, feature count, median and appendix numbering;
• the ProSurf cache key covers every atom;
• AF3 model ranking and structure lookup are hardened;
• the af3.py and test-import leftovers are cleaned up.

🤖 Generated with Claude Code