Module molcrawl.compounds.utils.preprocessing

Functions

def get_invalid_smiles_stats()

Expand source code

def get_invalid_smiles_stats():
    """
    Get statistics for invalid SMILES

        Returns:
    tuple: (number of invalid SMILES, total number of SMILES, ineffectiveness, list of invalid examples)
    """
    if _total_smiles_count == 0:
        return 0, 0, 0.0, []
    invalid_rate = (_invalid_smiles_count / _total_smiles_count) * 100
    return _invalid_smiles_count, _total_smiles_count, invalid_rate, _invalid_smiles_examples

Get statistics for invalid SMILES

Returns:

tuple: (number of invalid SMILES, total number of SMILES, ineffectiveness, list of invalid examples)

def prepare_scaffolds(smiles: str)

Expand source code

def prepare_scaffolds(smiles: str):
    """
        Prepare the scaffolds of a molecule.

        Args:
    smiles: SMILES string

        Returns:
    str: scaffold SMILES string, empty string if invalid

        Note:
    Large databases such as ZINC20 may contain invalid SMILES for the following reasons:
    1. Notation problems for ionic structures such as quaternary ammonium (N+)
    2. Conversion error from different formats
    3. Expression of special stereochemistry
    4. Automatic processing error when creating database

    These are usually within an acceptable range of a few percent of the total database.
    """
    global _invalid_smiles_count, _total_smiles_count, _invalid_smiles_examples
    _total_smiles_count += 1

    if smiles == "." or not smiles:
        _invalid_smiles_count += 1
        if len(_invalid_smiles_examples) < 10:
            _invalid_smiles_examples.append(("empty or dot", smiles))
        return ""

    if Chem is None or GetScaffoldForMol is None:
        raise ModuleNotFoundError("rdkit is required to prepare scaffolds")

    try:
        molecule = Chem.MolFromSmiles(smiles)
        if molecule is None:
            _invalid_smiles_count += 1
            # save first 10 invalid examples
            if len(_invalid_smiles_examples) < 10:
                _invalid_smiles_examples.append(("parse_failed", smiles[:100]))

            # Log statistics every 1000 items
            if _invalid_smiles_count % 1000 == 0:
                invalid_count, total_count, invalid_rate, examples = get_invalid_smiles_stats()
                logger.warning(f"Invalid SMILES detected: {invalid_count}/{total_count} ({invalid_rate:.2f}%)")
            return ""

        scaffold = GetScaffoldForMol(molecule)
        scaffold_smiles = Chem.MolToSmiles(scaffold)
        return scaffold_smiles
    except Exception as e:
        _invalid_smiles_count += 1
        if len(_invalid_smiles_examples) < 10:
            _invalid_smiles_examples.append(("exception", f"{smiles[:100]} | Error: {str(e)[:50]}"))
        logger.debug(f"Error processing SMILES '{smiles[:50]}...': {e}")
        return ""

Prepare the scaffolds of a molecule.

Args:

smiles: SMILES string

Returns:

str: scaffold SMILES string, empty string if invalid

Note:

Large databases such as ZINC20 may contain invalid SMILES for the following reasons: 1. Notation problems for ionic structures such as quaternary ammonium (N+) 2. Conversion error from different formats 3. Expression of special stereochemistry 4. Automatic processing error when creating database

These are usually within an acceptable range of a few percent of the total database.