feat(bbb): add Morgan fingerprint extraction with shape/dtype guarantees

src/pipelines/bbb_pipeline.py

@@ -12,7 +12,10 @@ from __future__ import annotations
 
 import math
 
+import numpy as np
 from rdkit import Chem, RDLogger
+from rdkit.Chem import AllChem
+from rdkit.DataStructs import ConvertToNumpyArray
 
 from src.core.logger import get_logger
 
@@ -42,3 +45,34 @@ def is_valid_smiles(smiles: str | float | None) -> bool:
     if not isinstance(smiles, str) or not smiles.strip():
         return False
     return Chem.MolFromSmiles(smiles) is not None
+
+
+def compute_morgan_fingerprint(
+    smiles: str,
+    n_bits: int = 2048,
+    radius: int = 2,
+) -> np.ndarray:
+    """Compute the Morgan (ECFP-like) circular fingerprint for a SMILES.
+
+    Args:
+        smiles: A SMILES string already known to be valid. Pass through
+            `is_valid_smiles` first if the source is untrusted.
+        n_bits: Length of the bit vector. 2048 is the de-facto default
+            for downstream scikit-learn classifiers.
+        radius: Morgan radius (2 ≈ ECFP4).
+
+    Returns:
+        A 1-D `np.ndarray` of length `n_bits` and dtype `uint8`, where
+        each element is 0 or 1.
+
+    Raises:
+        ValueError: if `smiles` cannot be parsed by RDKit.
+    """
+    mol = Chem.MolFromSmiles(smiles)
+    if mol is None:
+        raise ValueError(f"invalid SMILES: {smiles!r}")
+
+    bit_vect = AllChem.GetMorganFingerprintAsBitVect(mol, radius=radius, nBits=n_bits)
+    arr = np.zeros((n_bits,), dtype=np.uint8)
+    ConvertToNumpyArray(bit_vect, arr)
+    return arr

tests/pipelines/test_bbb_pipeline.py

@@ -31,3 +31,29 @@ class TestIsValidSmiles:
     def test_rejects_nan(self) -> None:
         import math
         assert is_valid_smiles(math.nan) is False
+
+
+import numpy as np
+
+from src.pipelines.bbb_pipeline import compute_morgan_fingerprint
+
+
+class TestComputeMorganFingerprint:
+    def test_returns_numpy_array_of_correct_length(self) -> None:
+        fp = compute_morgan_fingerprint("CCCO", n_bits=2048, radius=2)
+        assert isinstance(fp, np.ndarray)
+        assert fp.shape == (2048,)
+        assert fp.dtype == np.uint8
+
+    def test_only_zero_or_one(self) -> None:
+        fp = compute_morgan_fingerprint("c1ccccc1", n_bits=1024, radius=2)
+        assert set(np.unique(fp).tolist()).issubset({0, 1})
+
+    def test_different_molecules_yield_different_fingerprints(self) -> None:
+        fp_a = compute_morgan_fingerprint("CCCO", n_bits=2048, radius=2)
+        fp_b = compute_morgan_fingerprint("c1ccccc1", n_bits=2048, radius=2)
+        assert not np.array_equal(fp_a, fp_b)
+
+    def test_invalid_smiles_raises_value_error(self) -> None:
+        with pytest.raises(ValueError, match="invalid SMILES"):
+            compute_morgan_fingerprint("not_a_smiles", n_bits=2048, radius=2)