Skip to content

SKILL.md

Latest commit

 

History

History
200 lines (142 loc) · 6.2 KB

SKILL.md

File metadata and controls

200 lines (142 loc) · 6.2 KB
name bio-molecular-descriptors
description Calculates molecular descriptors and fingerprints using RDKit. Computes Morgan fingerprints (ECFP), MACCS keys, Lipinski properties, QED drug-likeness, TPSA, and 3D conformer descriptors. Use when featurizing molecules for machine learning or filtering by drug-likeness criteria.
tool_type python
primary_tool RDKit

Version Compatibility

Reference examples tested with: RDKit 2024.03+, numpy 1.26+, pandas 2.2+

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.

Molecular Descriptors

"Calculate molecular fingerprints for my compound library" → Compute structural fingerprints (Morgan/ECFP, MACCS keys) and physicochemical descriptors (Lipinski, QED, TPSA) for molecules, producing feature vectors for similarity analysis or ML models.

  • Python: AllChem.GetMorganFingerprintAsBitVect(), Descriptors.MolWt(), QED.qed() (RDKit)

Calculate fingerprints and physicochemical properties for molecules.

Morgan Fingerprints (ECFP)

Goal: Generate circular fingerprints that encode local chemical environments for similarity searching and ML models.

Approach: Use GetMorganFingerprintAsBitVect with a chosen radius (2 for ECFP4, 3 for ECFP6) and bit length, optionally including chirality information.

from rdkit import Chem
from rdkit.Chem import AllChem

mol = Chem.MolFromSmiles('CCO')

# ECFP4 = radius 2 (diameter = 2 * radius + 2 = 6)
# ECFP6 = radius 3 (diameter = 8)
ecfp4 = AllChem.GetMorganFingerprintAsBitVect(mol, radius=2, nBits=2048)
ecfp6 = AllChem.GetMorganFingerprintAsBitVect(mol, radius=3, nBits=2048)

# With stereochemistry information
ecfp4_chiral = AllChem.GetMorganFingerprintAsBitVect(
    mol, radius=2, nBits=2048, useChirality=True
)

# As count vector (for some ML methods)
ecfp4_counts = AllChem.GetMorganFingerprint(mol, radius=2)

# Convert to numpy array
import numpy as np
fp_array = np.array(ecfp4)

MACCS Keys

from rdkit.Chem import MACCSkeys

maccs = MACCSkeys.GenMACCSKeys(mol)  # 167 bits

# As numpy array
maccs_array = np.array(maccs)

Lipinski Properties

from rdkit import Chem
from rdkit.Chem import Descriptors, Lipinski

mol = Chem.MolFromSmiles('CCO')

# Lipinski Rule of 5 properties
mw = Descriptors.MolWt(mol)           # Molecular weight (<=500)
logp = Descriptors.MolLogP(mol)       # LogP (<=5)
hbd = Lipinski.NumHDonors(mol)        # H-bond donors (<=5)
hba = Lipinski.NumHAcceptors(mol)     # H-bond acceptors (<=10)

# Check Lipinski compliance
def passes_lipinski(mol):
    '''Check Lipinski Rule of 5 compliance.'''
    return (
        Descriptors.MolWt(mol) <= 500 and
        Descriptors.MolLogP(mol) <= 5 and
        Lipinski.NumHDonors(mol) <= 5 and
        Lipinski.NumHAcceptors(mol) <= 10
    )

# Additional properties
tpsa = Descriptors.TPSA(mol)          # Topological polar surface area
rotatable = Lipinski.NumRotatableBonds(mol)

QED Drug-Likeness

from rdkit.Chem.QED import qed

# QED score (0-1 scale, >0.5 generally drug-like)
qed_score = qed(mol)

# Weighted QED (default)
# Considers MW, LogP, TPSA, HBD, HBA, PSA, RotBonds, Aromatic rings

Complete Descriptor Set

Goal: Calculate all available RDKit molecular descriptors for feature-rich ML input.

Approach: Build a MolecularDescriptorCalculator from the full descriptor list and apply it to each molecule, producing a descriptor DataFrame.

from rdkit.Chem import Descriptors
from rdkit.ML.Descriptors import MoleculeDescriptors

# Get all available descriptor names
descriptor_names = [d[0] for d in Descriptors.descList]

# Create descriptor calculator
calculator = MoleculeDescriptors.MolecularDescriptorCalculator(descriptor_names)

# Calculate for a molecule
descriptors = calculator.CalcDescriptors(mol)

# As DataFrame
import pandas as pd
desc_df = pd.DataFrame([descriptors], columns=descriptor_names)

3D Conformer Descriptors

Goal: Compute 3D shape descriptors (asphericity, eccentricity, radius of gyration) from molecular conformers.

Approach: Generate a 3D conformer with ETKDGv3, optimize geometry with MMFF, then calculate 3D descriptors from the conformer coordinates.

from rdkit import Chem
from rdkit.Chem import AllChem, Descriptors3D

mol = Chem.MolFromSmiles('CCO')
mol = Chem.AddHs(mol)

# Generate 3D conformer (ETKDGv3 is now default)
AllChem.EmbedMolecule(mol, AllChem.ETKDGv3())

# Optimize geometry
AllChem.MMFFOptimizeMolecule(mol)

# 3D descriptors (require conformer)
# Asphericity: 0 = sphere, 1 = rod
asphericity = Descriptors3D.Asphericity(mol)

# Eccentricity
eccentricity = Descriptors3D.Eccentricity(mol)

# Inertial shape factor
isf = Descriptors3D.InertialShapeFactor(mol)

# Radius of gyration
rog = Descriptors3D.RadiusOfGyration(mol)

Batch Descriptor Calculation

Goal: Calculate a standard set of descriptors across an entire compound library.

Approach: Iterate over molecules, compute selected descriptors for each, and collect results into a DataFrame.

def calculate_descriptors_batch(molecules, descriptor_names=None):
    '''Calculate descriptors for multiple molecules.'''
    if descriptor_names is None:
        descriptor_names = ['MolWt', 'MolLogP', 'TPSA', 'NumHDonors',
                           'NumHAcceptors', 'NumRotatableBonds', 'qed']

    results = []
    for mol in molecules:
        if mol is None:
            results.append({d: None for d in descriptor_names})
            continue

        row = {}
        for name in descriptor_names:
            if name == 'qed':
                from rdkit.Chem.QED import qed
                row[name] = qed(mol)
            else:
                row[name] = getattr(Descriptors, name)(mol)
        results.append(row)

    return pd.DataFrame(results)

Related Skills

  • molecular-io - Load molecules for descriptor calculation
  • similarity-searching - Use fingerprints for similarity
  • admet-prediction - Predict ADMET from descriptors
  • machine-learning/biomarker-discovery - ML on molecular features