Fingerprint Utilities

Supported Fingerprints

The module supports several standard fingerprint types used in cheminformatics:

Fingerprint	Type	Description	Bits
ECFP	Structural	Extended Connectivity Fingerprint (Morgan)	Configurable
MACCS	Structural	MACCS Keys fingerprint	166
RDKit	Structural	RDKit topological fingerprint	2048
Gobbi2D	Pharmacophore	Gobbi pharmacophore 2D fingerprint	Variable
Avalon	Structural	Avalon fingerprint	Variable

ECFP Parameters

ECFP (Extended Connectivity Fingerprint, also known as Morgan fingerprint) fingerprints are specified as:

• Format: "ecfp{diameter}-{nbits}"
• Examples: "ecfp2-1024", "ecfp4-1024", "ecfp6-2048"
• Diameter: Topological diameter (internally converted to radius = diameter/2)
• Bits: Number of bits in the fingerprint (256, 512, 1024, 2048, etc.)

Tanimoto Similarity

The module uses Tanimoto (Jaccard) similarity to measure molecular similarity: - Range: 0.0 (completely dissimilar) to 1.0 (identical) - Formula: |A ∩ B| / |A ∪ B| - Suitable for bit vectors and sparse representations

Function Reference

Molecular fingerprint utilities for similarity calculations.

This module provides utilities for computing molecular fingerprints and similarity matrices. It supports multiple fingerprint types (ECFP, MACCS, RDKit, Gobbi2D, Avalon) and computes Tanimoto similarity between molecules for use in diversity-aware metrics.

fp_name_to_fn(fp_name)

Convert a fingerprint name to a function that generates that fingerprint.

This function returns a callable that takes an RDKit Mol object and returns the corresponding fingerprint. It supports several standard fingerprint types used in cheminformatics.

Parameters:

Name	Type	Description	Default
fp_name	str	Name of the fingerprint to use. Supported options: - "ecfp{diameter}-{nbits}": Extended Connectivity Fingerprint (Morgan). Examples: "ecfp2-1024", "ecfp4-1024", "ecfp6-2048". - "maccs": MACCS Keys fingerprint (166 bit). - "rdkit": RDKit fingerprint. - "Gobbi2d": Gobbi pharmacophore 2D fingerprint. - "Avalon": Avalon fingerprint.	required

Returns:

Type	Description
Callable[[Mol], ExplicitBitVect]	A callable function that takes an RDKit Mol object and returns
Callable[[Mol], ExplicitBitVect]	an ExplicitBitVect fingerprint.

Raises:

Type	Description
ValueError	If the fingerprint name is not recognized or has invalid format.
AssertionError	If the ECFP parameters don't match expected format.

Example

from mol_gen_docking.evaluation.fingeprints_utils import fp_name_to_fn
from rdkit import Chem

# Get the fingerprint function
fp_fn = fp_name_to_fn("ecfp4-1024")

# Generate fingerprints for molecules
mol = Chem.MolFromSmiles("c1ccccc1")
fp = fp_fn(mol)

Notes

• ECFP diameter is specified as topological diameter (divided by 2 internally to get the radius for Morgan fingerprints)
• All returned fingerprints are explicit bit vectors for similarity calculations

Source code in mol_gen_docking/evaluation/fingeprints_utils.py

def fp_name_to_fn(
    fp_name: str,
) -> Callable[[Chem.Mol], DataStructs.cDataStructs.ExplicitBitVect]:
    """Convert a fingerprint name to a function that generates that fingerprint.

    This function returns a callable that takes an RDKit Mol object and returns
    the corresponding fingerprint. It supports several standard fingerprint types
    used in cheminformatics.

    Args:
        fp_name: Name of the fingerprint to use. Supported options:
            - "ecfp{diameter}-{nbits}": Extended Connectivity Fingerprint (Morgan).
              Examples: "ecfp2-1024", "ecfp4-1024", "ecfp6-2048".
            - "maccs": MACCS Keys fingerprint (166 bit).
            - "rdkit": RDKit fingerprint.
            - "Gobbi2d": Gobbi pharmacophore 2D fingerprint.
            - "Avalon": Avalon fingerprint.

    Returns:
        A callable function that takes an RDKit Mol object and returns
        an ExplicitBitVect fingerprint.

    Raises:
        ValueError: If the fingerprint name is not recognized or has invalid format.
        AssertionError: If the ECFP parameters don't match expected format.

    Example:
        ```python
        from mol_gen_docking.evaluation.fingeprints_utils import fp_name_to_fn
        from rdkit import Chem

        # Get the fingerprint function
        fp_fn = fp_name_to_fn("ecfp4-1024")

        # Generate fingerprints for molecules
        mol = Chem.MolFromSmiles("c1ccccc1")
        fp = fp_fn(mol)
        ```

    Notes:
        - ECFP diameter is specified as topological diameter (divided by 2 internally
          to get the radius for Morgan fingerprints)
        - All returned fingerprints are explicit bit vectors for similarity calculations
    """

    if fp_name.startswith("ecfp"):
        d = int(fp_name[4])
        n_bits = int(fp_name.split("-")[1])
        assert fp_name == f"ecfp{d}-{n_bits}", f"Invalid fingerprint name: {fp_name}"

        def fp_fn(mol: Chem.Mol) -> DataStructs.cDataStructs.ExplicitBitVect:
            return AllChem.GetMorganFingerprintAsBitVect(mol, d // 2, n_bits)

        return fp_fn
    elif fp_name == "maccs":

        def fp_fn(mol: Chem.Mol) -> DataStructs.cDataStructs.ExplicitBitVect:
            return Chem.rdMolDescriptors.GetMACCSKeysFingerprint(mol)

        return fp_fn
    elif fp_name == "rdkit":

        def fp_fn(mol: Chem.Mol) -> DataStructs.cDataStructs.ExplicitBitVect:
            return Chem.RDKFingerprint(mol)

        return fp_fn
    elif fp_name == "Gobbi2d":
        from rdkit.Chem.Pharm2D import Generate, Gobbi_Pharm2D

        def fp_fn(mol: Chem.Mol) -> DataStructs.cDataStructs.ExplicitBitVect:
            return Generate.Gen2DFingerprint(mol, Gobbi_Pharm2D.factory)

        return fp_fn
    elif fp_name == "Avalon":
        from rdkit.Avalon import pyAvalonTools

        def fp_fn(mol: Chem.Mol) -> DataStructs.cDataStructs.ExplicitBitVect:
            return pyAvalonTools.GetAvalonFP(mol)

        return fp_fn
    else:
        raise ValueError(f"Unknown fingerprint name: {fp_name}")

get_sim_matrix(mols, fingerprint_name='ecfp4-1024')

Compute a pairwise similarity matrix for a list of molecules.

This function computes Tanimoto similarities between all pairs of molecules using the specified fingerprint method. The result is a condensed distance matrix (upper triangle only) in NumPy array format.

Parameters:

Name	Type	Description	Default
mols	list[Mol]	List of RDKit Mol objects.	required
fingerprint_name	str	Name of the fingerprint to use for similarity calculation. Default is "ecfp4-1024" (ECFP with diameter 4 and 1024 bits). See fp_name_to_fn for supported options.	'ecfp4-1024'

Returns:

Type	Description
ndarray[float]	A 1D NumPy array containing the upper triangle of the pairwise similarity
ndarray[float]	matrix. The array is in condensed form (as used by scipy.spatial.distance.squareform).
ndarray[float]	For n molecules, the length is n*(n-1)/2.

Example

from mol_gen_docking.evaluation.fingeprints_utils import get_sim_matrix
from rdkit import Chem

smiles = ["c1ccccc1", "CC(C)Cc1ccc(cc1)C(C)C(O)=O", "CCO"]
mols = [Chem.MolFromSmiles(smi) for smi in smiles]

sim_matrix = get_sim_matrix(mols, fingerprint_name="ecfp4-1024")
print(f"Shape: {sim_matrix.shape}")  # (3,) for 3 molecules

Notes

• The returned array is in condensed form. Use scipy.spatial.distance.squareform to convert to a full square similarity matrix.
• Similarity values range from 0.0 (completely dissimilar) to 1.0 (identical)
• The matrix is symmetric, so only the upper triangle is computed

Source code in mol_gen_docking/evaluation/fingeprints_utils.py

def get_sim_matrix(
    mols: list[Chem.Mol],
    fingerprint_name: str = "ecfp4-1024",
) -> np.ndarray[float]:
    """Compute a pairwise similarity matrix for a list of molecules.

    This function computes Tanimoto similarities between all pairs of molecules
    using the specified fingerprint method. The result is a condensed distance
    matrix (upper triangle only) in NumPy array format.

    Args:
        mols: List of RDKit Mol objects.
        fingerprint_name: Name of the fingerprint to use for similarity calculation.
            Default is "ecfp4-1024" (ECFP with diameter 4 and 1024 bits).
            See fp_name_to_fn for supported options.

    Returns:
        A 1D NumPy array containing the upper triangle of the pairwise similarity
        matrix. The array is in condensed form (as used by scipy.spatial.distance.squareform).
        For n molecules, the length is n*(n-1)/2.

    Example:
        ```python
        from mol_gen_docking.evaluation.fingeprints_utils import get_sim_matrix
        from rdkit import Chem

        smiles = ["c1ccccc1", "CC(C)Cc1ccc(cc1)C(C)C(O)=O", "CCO"]
        mols = [Chem.MolFromSmiles(smi) for smi in smiles]

        sim_matrix = get_sim_matrix(mols, fingerprint_name="ecfp4-1024")
        print(f"Shape: {sim_matrix.shape}")  # (3,) for 3 molecules
        ```

    Notes:
        - The returned array is in condensed form. Use scipy.spatial.distance.squareform
          to convert to a full square similarity matrix.
        - Similarity values range from 0.0 (completely dissimilar) to 1.0 (identical)
        - The matrix is symmetric, so only the upper triangle is computed
    """
    fp_fn = fp_name_to_fn(fingerprint_name)
    fps = [fp_fn(mol) for mol in mols]
    sim_mat = [
        np.array(DataStructs.BulkTanimotoSimilarity(fp, fps[i + 1 :]))
        for i, fp in enumerate(fps[:-1])
    ]
    matrix: np.ndarray[float] = np.concatenate(sim_mat)
    return matrix