[vinvomero]

Addresses the EC/GO portion of #96.

Problem

annotate() merges functional annotations with exact-key checks:

if "ec_number" in fa:
    n["ec_number"][i] = fa["ec_number"]
if "go_terms" in fa:
    n["go_terms"][i] = fa["go_terms"]

but eggNOG / funannotate2-addons write the EC number as EC_number, and InterProScan writes GO terms as go_term (singular). Those annotations were loaded and then silently dropped. In the #96 reporter's data, a local patch for these keys recovered 5,296 GO terms.

Fix

Adds utilities.resolve_annotation_key(fa, canonical_key), which finds the key present in an annotation dict for ec_number / go_terms, tolerant of those known variants. A key that actually carries data is preferred, so an empty canonical key never hides a populated variant; when both carry data the canonical key wins, so existing canonical-key inputs behave exactly as before. The EC/GO merge in annotate() now goes through it.

Scope

This covers the EC and GO keys from #96. The InterPro/SignalP summary counts the reporter also mentions are computed in the gfftk dependency (gfftk/stats.py), so they aren't changed here.

Tests

New tests/unit/test_utilities_annotation.py (9 tests): canonical keys, the EC_number / go_term variants, a populated variant winning over an empty canonical key, the canonical key winning when both are populated, absent keys, unknown keys matching only themselves, and a present-but-None value being preserved.

Full unit suite run locally on top of current main: 205 passed vs 196 on unmodified main. The difference is the 9 new tests, with no regressions. (Run on Windows, where annorefine and gapmm2 don't install, so test_predict*.py didn't run locally, and one tmpdir test fails identically on main because it expects a /tmp path. CI here should cover those.)

🤖 Generated with Claude Code