openvax / topiary / 33426610623

93%

push

github

Add read-level evidence and per-field knownness to ProteinFragment (#102) (#221)

First half of the multi-source fragment work, from vaxrank's consumer
requirements. isovar integration is deliberately not here.

Four RNA read-count fields, none derivable from the aggregate expression
the fragment already carried, and one of which -- reads supporting this
assembled protein sequence rather than merely the variant allele -- is a
distinction only the assembler can make.

None means unknown and is not 0. A source with no read data leaves them
None; a source that looked and found nothing sets 0. Collapsing those
would let a consumer read "no RNA support" out of "this source cannot
answer". The distinction survives a TSV round trip, since one that does
not serialize is decorative.

field_provenance says how real a populated field is: measured,
approximated, or synthesized. That covers what a bare None cannot -- a
LENS or pVACseq read count is real but estimated, and a placeholder
ref/alt invented because the source supplied none has a value that means
nothing, which anything doing variant effect annotation must refuse
rather than compute on. Accessors (is_known, provenance_of,
is_approximate, is_usable_as_biology) rather than the raw dict. A
provenance entry naming a nonexistent field, or an unknown label, is
refused: a typo would sit inert and quietly stop protecting the field it
was written for.

Also fixes ProteinFragment.from_dict, which hardcoded its field list and
so rejected every field added after that list was written. It now derives
the set from the dataclass and cannot drift again -- these new fields
would have been the first casualties.

18 tests, including the degenerate-pVACseq conformance case vaxrank
proposed and an explicit "a consumer need not branch on source_type".
Version 5.32.0.

Claude-Session: https://claude.ai/code/session_01BAUqjqDDPXaTdtUDJqapJG

39 of 40 new or added lines in 3 files covered. (97.5%)

5766 of 6326 relevant lines covered (91.15%)

2.73 hits per line