Mol and atommap-annotated Mol produces different ring molecules

[loryruta]

Hello,
first of all I'd like to precise that I have very little knowledge about chemistry (I'm a programmer/ML engineer).

I want to decompose a molecule into rings and bonds for constructing a Motif vocabulary (similar to the one cited by many SOTA techniques).

I have these two SMILES that represent the same molecule:

1. O=S1OCc2ccccc21
2. [cH:4]1[cH:5][cH:6][cH:7][c:8]2[c:9]1[S:10](=[O:11])[O:12][CH2:13]2
   The second one is annotated with atommap.

Now, after extracting bonds and rings I obtain one differing sub-molecule (the first):

1. C1COSC1 O=S c1ccccc1
2. C1=CSOC1 O=S c1ccccc1

The code I'm using for extracting rings is:

def mol_from_smiles(smiles: str, sanitize=True, kekulize=True) -> Chem.Mol:
    mol = Chem.MolFromSmiles(smiles, sanitize=sanitize)
    if mol is None:
        raise Exception(f"Invalid SMILES string: \"{smiles}\"")
    if kekulize:
        Chem.Kekulize(mol)
    return mol

def extract_mol_fragment(mol: Chem.Mol, atom_indices: Set[int]) -> Chem.Mol:
    """ Extracts a fragment (subset) of the input molecule keeping only the specified atoms.
    The chemical validity of the output fragment is ensured by incrementally building it with RWMol.
    """

    frag_mol = Chem.RWMol()

    old_to_new_atom_idx = {}

    for i in atom_indices:
        new_idx = frag_mol.AddAtom(copy_atom(mol.GetAtomWithIdx(i)))
        old_to_new_atom_idx[i] = new_idx

    for bond in mol.GetBonds():
        u = bond.GetBeginAtomIdx()
        v = bond.GetEndAtomIdx()
        if u in atom_indices and v in atom_indices:
            frag_mol.AddBond(
                old_to_new_atom_idx[u],
                old_to_new_atom_idx[v],
                bond.GetBondType()
            )

    Chem.Kekulize(frag_mol)

    return frag_mol.GetMol()

def canon_smiles(smiles: str) -> str:
    smiles = Chem.MolToSmiles(mol_from_smiles(smiles), kekuleSmiles=True)  # Kekulize
    return Chem.CanonSmiles(smiles)

def extract_rings(mol_smiles: str) -> Set[str]:
    rings = set()
    for atom_indices in Chem.GetSymmSSSR(mol):
        ring_mol = extract_mol_fragment(mol, atom_indices)
        ring_smiles = Chem.MolToSmiles(ring_mol)
        rings.add(canon_smiles(ring_smiles))

Since I run the same extraction code on both 1. and 2. and the only different between those are the atom map annotations, I can't see why I'm getting different results. Can anyone help me to spot which could be the cause?