Metabolite → gene links: transporter links via KEGG BRITE ko02000

[sr320]

Three commits from the claude/recursing-varahamihira-f8dfa9 session that landed after #73 (metabolite_linked_convergence tier) had already merged, so they never reached main. They're moved to a fresh branch, with current main merged in (no conflicts).

Commits

1. 033d3c1 Add transporter metabolite → gene links via KEGG BRITE ko02000
   • KEGG BRITE ko02000 classifies transporter KOs into families but records no substrates. The KO → compound step is therefore curated in data/mappings/transporter_substrates.tsv: 43 rows, each with substrate_basis, a human ortholog and a note. For example: GLUT/SGLT → glucose, MCT/SMCT → lactate/pyruvate, NaDC/DIC → succinate, BGT1 → betaine, system A/ASCT/PAT → alanine.
   • The builder fetches br:ko02000, rejects curated KOs that aren't in it, and records each KO's BRITE family path.
   • Transporter links are always inferred. The hub cap is applied per link type, so transporter links never strip a compound's enzyme-link credit.
   • The link table gains link_basis and brite_family. Evidence cards show link type, basis and family, with a note on why transporter links rest on curation.
   • First local build: 191 enzyme + 58 transporter links. (S)-lactate, which has no L-LDH enzyme link in crg, now links through MCT/SMCT.
2. a6e1646 Demo: glucose ↔ simulated glut1 transporter link. Adds a fabricated glut1 gene (LOC105342003) to the demo crosswalk and GIGAS_HEAT01, plus a transporter row with the real GLUT1 KO K07299 and its BRITE path. The demo now shows both link types reaching metabolite_linked_convergence.
3. 8c9abe2 Demo: move GIGAS_GROW06 glucose off the adj. p = 0.05 boundary. Glucose sat exactly at padj 0.05, so the demo case depended on the gate using <=. Its p-value goes from 0.021 to 0.006 and its padj from 0.05 to 0.02 (simulated data). A test checks that every demo link-convergence feature has adj. p ≤ 0.04.

Checked locally (after merging current main)

• ruff clean; pytest 730 passed
• CI demo pipeline: register → harmonize → meta-analyze → build-evidence-cards → top-candidates completes. The glucose evidence card shows the transporter link.

All demo changes are to simulated: true data and the synthetic demo crosswalk; no real study is affected.

🤖 Generated with Claude Code