Add MolFromInchiAndAuxInfo to restore original atom order from AuxInfo

[rodyarantes]

Summary

Adds MolFromInchiAndAuxInfo(inchi, auxinfo, ...) to inchi.py — the inverse of the existing MolToInchiAndAuxInfo. It reconstructs a molecule from an InChI string and its AuxInfo, restoring:

• Original atom ordering via the /N: (atom numbering) layer
• 2D/3D coordinates via the /rC: (coordinate) layer

This enables lossless round-tripping of molecules through InChI when the AuxInfo is preserved, which is useful for workflows where atom identity and spatial layout must survive serialization.

New public API

• MolFromInchiAndAuxInfo(inchi, auxinfo, sanitize=True, removeHs=True, logLevel=None, treatWarningAsError=False) → Chem.rdchem.Mol

Internal helpers added

• _parse_auxinfo_coordinates(auxinfo) — parses /rC: layer into [(x,y,z), ...]
• _parse_auxinfo_atom_order(auxinfo) — parses /N: layer into 0-based index list
• _attach_conformer(mol, coords, is_3d) — adds a conformer from parsed coordinates
• _build_inverse_permutation(atom_order, size) — builds the permutation array for RenumberAtoms

Edge cases handled

• None or empty AuxInfo → falls back to MolFromInchi behavior
• Invalid InChI → returns None
• Multi-fragment molecules (semicolons in /N: layer)
• Trailing semicolons in /rC: (standard InChI output format)
• All-zero coordinates → no conformer attached
• removeHs=True vs False atom count mismatches

Usage examples

Basic round-trip preserving atom order

from rdkit import Chem
from rdkit.Chem import MolFromSmiles, MolToSmiles
from rdkit.Chem import MolToInchiAndAuxInfo, MolFromInchiAndAuxInfo

mol = MolFromSmiles('c1cc(O)ccc1N')
inchi, aux = MolToInchiAndAuxInfo(mol)

# Reconstruct with original atom ordering restored
mol2 = MolFromInchiAndAuxInfo(inchi, aux)

# Atom order is preserved
orig_atoms = [a.GetSymbol() for a in mol.GetAtoms()]
new_atoms = [a.GetSymbol() for a in mol2.GetAtoms()]
assert orig_atoms == new_atoms

Round-trip with coordinate preservation

from rdkit.Chem import MolFromMolBlock, MolToInchiAndAuxInfo, MolFromInchiAndAuxInfo

mol = MolFromMolBlock(mol_block)  # molecule with 2D/3D coords
inchi, aux = MolToInchiAndAuxInfo(mol)

mol2 = MolFromInchiAndAuxInfo(inchi, aux)
assert mol2.GetNumConformers() == 1  # coordinates restored

# Coordinates match the original
for i in range(mol.GetNumAtoms()):
    orig = mol.GetConformer().GetAtomPosition(i)
    restored = mol2.GetConformer().GetAtomPosition(i)
    assert abs(orig.x - restored.x) < 1e-4
    assert abs(orig.y - restored.y) < 1e-4

Graceful degradation

# Works fine with no AuxInfo — just returns MolFromInchi result
mol = MolFromInchiAndAuxInfo('InChI=1S/CH4/h1H4', None)
assert mol is not None

# Returns None for invalid InChI
mol = MolFromInchiAndAuxInfo('not_an_inchi', '')
assert mol is None

Test plan

Tests are in rdkit/Chem/UnitTestInchi.py under TestMolFromInchiAndAuxInfo:

• test0RoundTripAtomOrder — atom ordering preserved for multiple SMILES
• test1StereoPreservation — stereochemistry preserved through round-trip
• test2NoneAuxInfo / test3EmptyAuxInfo — graceful fallback with missing AuxInfo
• test4InvalidInchi — returns None for invalid InChI
• test5MultiFragmentAuxInfo — handles semicolon-separated fragments
• test6CoordinateRestoration — 2D conformer restored, trailing semicolons handled
• test7EmptyCoordinates / test7bAllZeroCoordinates — no conformer for empty/zero coords
• test8CoordinateAtomOrderMatch — per-atom symbol + coordinate verification

To run:

python -m pytest rdkit/Chem/UnitTestInchi.py::TestMolFromInchiAndAuxInfo -v

[greglandrum]

Thanks for the contribution @rodyarantes!

I don't really know the auxinfo format, and I didn't dig into the regexes (:shudder:) but it looks like your tests are covering things nicely.

One small suggestion to simplify a test, but otherwise this looks good!

[greglandrum]

LGTM

[greglandrum]

Thanks @rodyarantes!