fix(predict): rename Tox21 models dict — it clobbered the structural-alert SMARTS (#61)

I introduced a name collision in #60: the new Tox21 models dict was named `_TOX`,
the same name as the existing structural-alert SMARTS dict. So `_tox_alerts()`
iterated RandomForest objects instead of SMARTS patterns — silently emptying the
structural-alert safety screen when no tox models were present (the CI case, which
is why CI missed it) and crashing predict() when they were (surfaced once the
demo box had trained tox_models/).

Fix: the tox models dict is now `_TOX_MODELS`; the structural-alert `_TOX` is intact.
Adds a regression test (nitrobenzene must trip the 'aromatic nitro' alert). Verified
end-to-end with the populated GRAS + properties tables: menthol BP 215 °C + in-GRAS,
nitrobenzene not-in-GRAS + nitro alert, tox heads loaded. 21 tests pass.

Signed-off-by: Austin L. <86896075+rvnminers-A-and-N@users.noreply.github.com>

Author: rvnminers-A-and-N

tests/test_predict.py

@@ -29,3 +29,11 @@ def test_predict_includes_tox_screen():
     out = predict.predict("CCO")
     assert "tox_screen" in out["safety"]
     assert isinstance(out["safety"]["tox_screen"]["available"], bool)
+
+
+def test_safety_structural_alert_fires():
+    # regression: a nitro aromatic must trip the structural tox-alert screen. (The tox-models
+    # dict was accidentally named _TOX, clobbering the structural-alert SMARTS dict; this
+    # silently emptied structural_alerts with no models and crashed with them.)
+    out = predict.predict("O=[N+]([O-])c1ccccc1")  # nitrobenzene
+    assert out["safety"]["structural_alerts"]  # non-empty

training/predict.py

@@ -187,11 +187,11 @@ def _ik1(*smiles):
 
 # Caution-only toxicity-assay heads (Tox21, public domain). Loaded if trained.
 # INDICATIVE in-vitro signals — never a toxicity determination.
-_TOX = {}
+_TOX_MODELS = {}
 _TOX_DIR = Path("tox_models")
 if _TOX_DIR.exists():
     for p in _TOX_DIR.glob("*_rf.joblib"):
-        _TOX[p.stem.replace("_rf", "")] = joblib.load(p)
+        _TOX_MODELS[p.stem.replace("_rf", "")] = joblib.load(p)
 
 # Known-label lookup: ground truth for molecules we actually have data on. This
 # is how the salty/sour data works as a FLAG without a model — if a queried
@@ -623,12 +623,12 @@ def predict_aroma(smiles, top_k=8):
 def predict_tox(mol, threshold=0.5):
     """Caution-only in-vitro tox-assay activity (Tox21 models). INDICATIVE flags for
     review — NEVER a toxicity/safety determination. Honest/empty if heads untrained."""
-    if not _TOX:
+    if not _TOX_MODELS:
         return {"available": False,
                 "note": "tox heads not trained — run train_tox.py (Tox21, public domain)"}
     x = _fp(mol)
     assays = []
-    for name, clf in sorted(_TOX.items()):
+    for name, clf in sorted(_TOX_MODELS.items()):
         p = round(float(clf.predict_proba(x)[0, 1]), 3)
         assays.append({"assay": name, "meaning": _TOX_MEANING.get(name, name), "probability": p})
     flags = [a["assay"] for a in assays if a["probability"] >= threshold]