Skip to content

feat: substitution search + build-status & OpenPOM-scope docs - #46

Merged
rvnminers-A-and-N merged 1 commit into
mainfrom
feat/substitution-and-build-status
Jun 27, 2026
Merged

rvnminers-A-and-N merged 1 commit into
mainfrom
feat/substitution-and-build-status

Conversation

@rvnminers-A-and-N

Copy link
Copy Markdown
Collaborator

What

Substitution search (#22 core). Adds substitute() to predict.py β€” Tanimoto/Morgan nearest-neighbor over the labeled molecule set, each neighbor returned with its known tastes. The reformulation / cost-down tool. Track-A implementation; Track B mirrors it as a pgvector ANN query over the same fingerprints (M6).

Spot-checked: glucose β†’ ribose/sugars (sim 0.94); vanillin β†’ ethyl ferulate / creosol / anisaldehyde.

Docs

  • BUILD-STATUS.md (new) β€” living inventory of what's built / built-but-needs-validation / stubbed / deferred, plus the product-vision staging (substitution now β†’ aroma-on-their-data pilot β†’ novel-molecule + retrosynthesis long-horizon).
  • AROMA.md β€” precise OpenPOM commercial scope: MIT code is usable, but the bundled GS-LF data is NonCommercial and weights trained on it are encumbered. A usable aroma model must be our own, trained from scratch on licensed (PMP 2001) or customer data.
  • ARCHITECTURE.md β€” aroma row now points to AROMA.md.

Notes

  • Validated substitute() against the real labeled set; ruff clean; imports OK.
  • No GS-LF data anywhere in the build β€” taste engine is ChemTastesDB (CC-BY) + SweetenersDB (MIT) only.

…POM scope

Add substitute() β€” Tanimoto/Morgan nearest-neighbor over the labeled
molecule set (issue #22 core); each neighbor carries its known tastes.
Track-A implementation; Track B mirrors it as a pgvector ANN query.

Docs:
- BUILD-STATUS.md: living inventory (built / needs-validation / stubbed /
  deferred) + product-vision staging (substitution now -> aroma-on-their-data
  pilot -> novel-molecule/retrosynthesis long-horizon).
- AROMA.md: precise OpenPOM commercial scope β€” MIT *code* is usable, but its
  GS-LF training data is NonCommercial and its weights are encumbered; a usable
  aroma model must be our own, trained from scratch on licensed/customer data.
- ARCHITECTURE.md: aroma row now points to AROMA.md.
@rvnminers-A-and-N
rvnminers-A-and-N merged commit 9f42364 into main Jun 27, 2026
2 checks passed
@rvnminers-A-and-N
rvnminers-A-and-N deleted the feat/substitution-and-build-status branch June 27, 2026 16:38
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant