openvax / topiary / 25882966575

90%

master: 90%

Pull #167

github

docs/pvacseq: drop broken concat example, fix kind_support reference

Three issues found in the last review pass of `docs/pvacseq.md`:

1. The re-scoring section's `concat([r, fresh_result])` example referenced
   an undefined variable (`fresh_result` vs `fresh` from the assignment
   above), and even with the name fixed `concat` doesn't accept bare
   DataFrames — `TopiaryPredictor.predict_*` returns a DataFrame, not
   a TopiaryResult.  Rather than widen `concat` (scope creep beyond
   the loader's needs) or wrap the DataFrame artificially, drop the
   downstream-composition example entirely.  The re-scoring section
   shows how to fresh-predict; users compose with pandas as needed.

2. `"class lives in `Metadata.kind_support`"` — `Metadata` (the
   comment-block dataclass) doesn't have a `kind_support` field.
   The correct location is `TopiaryPredictor.kind_support`.

3. The schema-mapping table's `(or Best MT IC50 Score)` parenthetical
   read as "either column is acceptable."  Clarified to tie the
   fallback to pVACseq's `--top-score-metric=Best` configuration;
   same for the WT percentile column (Median vs Corresponding).

Also tightens the peptide-dict construction to a dict-comprehension
that makes the intent explicit instead of a dict(zip(...)) expression
that collapses duplicates as a side effect.

Two follow-up issues filed for the design surface that came out of
the review but doesn't belong in this PR:

- #168 — Add `allele_set` column to faithfully store haplotype-mode
  predictions (schema layer)
- #169 — DSL: explicit per-`mhc_dependence` peptide-level projection
  (query layer)

Together those two unblock cross-source / cross-kind prediction
comparison cleanly; without them, validator / combine-predictions
helpers can only approximate.  Out of scope for the pVACseq loader.

268 of 301 new or added lines in 5 files covered. (89.04%)

4551 of 5064 relevant lines covered (89.87%)

2.7 hits per line