bio-similarity-searching

Version Compatibility

Reference examples tested with: RDKit 2024.03+

Before using code patterns, verify installed versions match. If versions differ:

• Python: pip show <package> then help(module.function) to check signatures

If code throws ImportError, AttributeError, or TypeError, introspect the installed

package and adapt the example to match the actual API rather than retrying.

Similarity Searching

"Find compounds similar to my query molecule" → Compute pairwise Tanimoto similarity on molecular fingerprints to rank a library by structural resemblance to a query, or cluster compounds by chemical similarity using Butina clustering.

• Python: DataStructs.TanimotoSimilarity(), Butina.ClusterData() (RDKit)

Find structurally similar molecules and cluster compound libraries.

Tanimoto Similarity

from rdkit import Chem, DataStructs
from rdkit.Chem import AllChem

# Generate fingerprints
mol1 = Chem.MolFromSmiles('CCO')
mol2 = Chem.MolFromSmiles('CCCO')

fp1 = AllChem.GetMorganFingerprintAsBitVect(mol1, radius=2, nBits=2048)
fp2 = AllChem.GetMorganFingerprintAsBitVect(mol2, radius=2, nBits=2048)

# Tanimoto similarity (0-1)
similarity = DataStructs.TanimotoSimilarity(fp1, fp2)
print(f'Tanimoto similarity: {similarity:.3f}')

Similarity Thresholds

| Threshold | Interpretation |

|-----------|----------------|

| > 0.85 | Very similar (likely same scaffold) |

| > 0.70 | Similar (likely related series) |

| > 0.50 | Moderate similarity |

| < 0.50 | Dissimilar |

Search Library Against Query

Goal: Find molecules structurally similar to a query compound within a library.

Approach: Generate fingerprints for the query and each library molecule, compute Tanimoto similarity, and return hits above a chosen threshold sorted by similarity.

from rdkit import Chem, DataStructs
from rdkit.Chem import AllChem

def find_similar_molecules(query_smiles, library, threshold=0.7, fp_type='ecfp4'):
    '''
    Find molecules similar to query in library.

    Args:
        query_smiles: Query molecule SMILES
        library: List of (smiles, name) tuples or SMILES list
        threshold: Minimum Tanimoto similarity
        fp_type: 'ecfp4', 'ecfp6', or 'maccs'
    '''
    query = Chem.MolFromSmiles(query_smiles)
    if query is None:
        raise ValueError('Invalid query SMILES')

    # Generate query fingerprint
    if fp_type == 'ecfp4':
        query_fp = AllChem.GetMorganFingerprintAsBitVect(query, 2, nBits=2048)
    elif fp_type == 'ecfp6':
        query_fp = AllChem.GetMorganFingerprintAsBitVect(query, 3, nBits=2048)
    else:  # maccs
        from rdkit.Chem import MACCSkeys
        query_fp = MACCSkeys.GenMACCSKeys(query)

    # Search library
    hits = []
    for item in library:
        smiles = item[0] if isinstance(item, tuple) else item
        name = item[1] if isinstance(item, tuple) and len(item) > 1 else smiles

        mol = Chem.MolFromSmiles(smiles)
        if mol is None:
            continue

        if fp_type == 'ecfp4':
            lib_fp = AllChem.GetMorganFingerprintAsBitVect(mol, 2, nBits=2048)
        elif fp_type == 'ecfp6':
            lib_fp = AllChem.GetMorganFingerprintAsBitVect(mol, 3, nBits=2048)
        else:
            lib_fp = MACCSkeys.GenMACCSKeys(mol)

        sim = DataStructs.TanimotoSimilarity(query_fp, lib_fp)
        if sim >= threshold:
            hits.append((smiles, name, sim))

    return sorted(hits, key=lambda x: x[2], reverse=True)

Bulk Similarity Search

from rdkit import DataStructs

def bulk_similarity_search(query_fp, library_fps, threshold=0.7):
    '''
    Fast similarity search using bulk operations.

    Args:
        query_fp: Query fingerprint
        library_fps: List of library fingerprints
        threshold: Minimum similarity
    '''
    # BulkTanimotoSimilarity is faster for large libraries
    similarities = DataStructs.BulkTanimotoSimilarity(query_fp, library_fps)

    hits = [(i, sim) for i, sim in enumerate(similarities) if sim >= threshold]
    return sorted(hits, key=lambda x: x[1], reverse=True)

Butina Clustering

Goal: Group a compound library into clusters of structurally similar molecules.

Approach: Compute an all-vs-all Tanimoto distance matrix from fingerprints and apply Taylor-Butina clustering with a distance cutoff.

from rdkit import Chem
from rdkit.ML.Cluster import Butina

def cluster_molecules(molecules, cutoff=0.4):
    '''
    Cluster molecules by Tanimoto similarity using Taylor-Butina algorithm.

    Args:
        molecules: List of RDKit mol objects
        cutoff: Distance cutoff (1 - similarity threshold)
               cutoff=0.4 means similarity threshold of 0.6
    '''
    # Generate fingerprints
    fps = [AllChem.GetMorganFingerprintAsBitVect(m, 2, nBits=2048)
           for m in molecules if m is not None]

    # Calculate distance matrix (upper triangle)
    n = len(fps)
    dists = []
    for i in range(1, n):
        sims = DataStructs.BulkTanimotoSimilarity(fps[i], fps[:i])
        dists.extend([1 - s for s in sims])

    # Cluster
    clusters = Butina.ClusterData(dists, n, cutoff, isDistData=True)

    return clusters

# Usage
# clusters = cluster_molecules(molecules, cutoff=0.3)  # 70% similarity
# print(f'Found {len(clusters)} clusters')
# for i, cluster in enumerate(clusters[:5]):
#     print(f'Cluster {i}: {len(cluster)} molecules')

Maximum Common Substructure

Goal: Identify the largest shared substructure across a set of molecules.

Approach: Use FindMCS with ring-matching constraints and a timeout to find the maximum common substructure as a SMARTS pattern.

from rdkit.Chem import rdFMCS

def find_mcs(molecules, timeout=60):
    '''Find maximum common substructure.'''
    mcs = rdFMCS.FindMCS(
        molecules,
        timeout=timeout,
        matchValences=False,
        ringMatchesRingOnly=True
    )
    return mcs.smartsString, mcs.numAtoms, mcs.numBonds

# Get MCS as molecule for visualization
mcs_smarts, n_atoms, n_bonds = find_mcs(molecules)
mcs_mol = Chem.MolFromSmarts(mcs_smarts)

Related Skills

• molecular-descriptors - Generate fingerprints for similarity
• substructure-search - Pattern-based searching
• molecular-io - Load molecules for searching