PaDEL-pywrapper 1.1.0

Python wrapper for the PaDEL descriptors

🧪 PaDEL Python Wrapper

A simple and reliable Python wrapper for calculating PaDEL molecular descriptors and fingerprints. This library takes care of installing a matching Java runtime, dispatching molecules to the bundled PaDEL executable, and collecting the results into a tidy pandas DataFrame — so you can stay in RDKit/pandas-land.

✨ Features

• 🧬 1875 descriptors & 12 fingerprint types — 1444 1D/2D and 431 3D descriptors, plus FP, ExtFP, EStateFP, GraphFP, MACCSFP, PubchemFP, SubFP(C), KRFP(C) and AP2DFP(C) fingerprints, straight from PaDEL-Descriptor.
• ☕ Zero Java setup — automatically downloads, caches and reuses a matching JRE on first use; nothing to install by hand.
• ⚡ Parallel by design — spread the work across multiple CPU cores with configurable njobs/chunksize, each worker running its own single-core-pinned JVM.
• 🛡️ Hang-proof — some 3D descriptors (e.g. WHIM) can fail to converge on degenerate geometries; an optional per-descriptor timeout guarantees you get NaN back instead of a frozen process.
• 🧯 Never silently misaligned — molecules that fail, get skipped, or lack hydrogens/conformers are handled explicitly and reported, never dropped without a trace.
• 📊 pandas-native output — results come back as a ready-to-use DataFrame, one row per molecule.
• 🔍 Rich metadata — inspect each descriptor's description, sub-components and 3D requirement programmatically.
• 🧩 Configurable fingerprints — tune bit size and search depth for the fingerprints that support it.

✍️ Copyright and Citation Notice

Olivier J. M. Béquignon is neither the copyright holder of PaDEL nor responsible for it. The work carried out here concerns:

• the Python wrapper,
• the ePaDEL executable,
• the extendedlibpadeldescriptor library.

Citing

If you use this wrapper in your research, please cite the original PaDEL publication in addition to this software package:

1. Original PaDEL paper:

   Yap, C.W. (2011), PaDEL-descriptor: An open source software to calculate molecular descriptors and fingerprints. Journal of Computational Chemistry, 32(7), 1466–1474. DOI: 10.1002/jcc.21707

2. This wrapper:

   Please refer to the repository at github.com/OlivierBeq/PaDEL_pywrapper for citation details.

📦 Installation

pip install padel-pywrapper

Or from source:

git clone https://github.com/OlivierBeq/PaDEL_pywrapper.git
pip install ./PaDEL_pywrapper

🛠️ Requirements

• Python 3.11+
• RDKit

💡 Usage

1D and 2D descriptors

Descriptors of the module PaDEL_pywrapper.descriptor can be computed as follows:

With pip:
from PaDEL_pywrapper import PaDEL
from PaDEL_pywrapper.descriptor import ALOGP, Crippen, FMF
from rdkit import Chem

smiles_list = [
# erlotinib
    "n1cnc(c2cc(c(cc12)OCCOC)OCCOC)Nc1cc(ccc1)C#C",
    # midecamycin
    "CCC(=O)O[C@@H]1CC(=O)O[C@@H](C/C=C/C=C/[C@@H]([C@@H](C[C@@H]([C@@H]([C@H]1OC)O[C@H]2[C@@H]([C@H]([C@@H]([C@H](O2)C)O[C@H]3C[C@@]([C@H]([C@@H](O3)C)OC(=O)CC)(C)O)N(C)C)O)CC=O)C)O)C",
    # selenofolate
    "C1=CC(=CC=C1C(=O)NC(CCC(=O)OCC[Se]C#N)C(=O)O)NCC2=CN=C3C(=N2)C(=O)NC(=N3)N",
]
mols = [Chem.MolFromSmiles(smiles) for smiles in smiles_list]

descriptors = [ALOGP, Crippen, FMF]

padel = PaDEL(descriptors)
print(padel.calculate(mols))

Instances of descriptors can be supplied as well:

descriptors = [ALOGP(), Crippen(), FMF()]

padel = PaDEL(descriptors)
print(padel.calculate(mols))

To calculate all possible descriptors, import the descriptors list from the PaDEL_pywrapper module directly:

from PaDEL_pywrapper import descriptors

padel = PaDEL(descriptors)
print(padel.calculate(mols))

3D descriptors

By default, the ignore_3D parameter is set to True, preventing any provided 3D descriptor from being calculated.

Should molecules with 3D coordinates be provided, one can turn on 3D descriptor calculation:

from rdkit.Chem import AllChem
from PaDEL_pywrapper.descriptor import WHIM

mols = [Chem.AddHs(mol) for mol in mols]
_ = [AllChem.EmbedMolecule(mol) for mol in mols]

descriptors_3d = [ALOGP, Crippen, FMF, WHIM]

padel = PaDEL(descriptors_3d, ignore_3D=False)
print(padel.calculate(mols))

⚠️ A warning is raised if molecules lack hydrogens. ⚠️ An exception is raised if molecules do not have 3D coordinates — but only when a 3D descriptor (like WHIM above) is actually requested; a purely 2D descriptor list computes 2D coordinates on the fly instead.

mol = Chem.MolFromSmiles('CCC')

padel = PaDEL(descriptors_3d, ignore_3D=False)
print(padel.calculate([mol]))
# ValueError: Cannot calculate descriptors for a conformer-less molecule

Some 3D descriptors (e.g. WHIM) rely on an eigenvalue decomposition that can fail to converge — and hang indefinitely — for molecules with degenerate 3D geometry (very few heavy atoms, near-linear/near-planar arrangements). 🛡️ Pass a timeout (in seconds) to guarantee that any descriptor exceeding it returns NaN for that molecule instead of blocking the whole batch:

padel = PaDEL(descriptors_3d, ignore_3D=False)
print(padel.calculate(mols, timeout=20))

Fingerprints

Fingerprints of the module PaDEL_pywrapper.descriptor can be computed as follows:

from PaDEL_pywrapper.descriptor import GraphOnlyFP

fp = GraphOnlyFP

padel = PaDEL([fp], ignore_3D=False)
print(padel.calculate(mols))

Custom parameter sets can be provided for some fingerprints:

fp = GraphOnlyFP(size=2048, searchDepth=8)

padel = PaDEL([fp], ignore_3D=False)
print(padel.calculate(mols))

⚡ Parallel processing

Speed things up by spreading molecules across several CPU cores. Each worker runs its own single-core-pinned JVM, so parallelism comes purely from the number of processes spawned — not from oversubscribing the host:

padel = PaDEL(descriptors)
print(padel.calculate(mols, njobs=8))

By default, molecules are auto-balanced evenly across njobs workers (chunksize=None), which minimizes JVM startup overhead while keeping every worker busy — the fastest setting for most workloads. A fixed chunksize can be provided instead if finer control is needed.

🔍 Details about descriptors

Details about each descriptor and fingerprint can be obtained as follows:

print(ALOGP.description)

print(GraphOnlyFP.description)

For full details about all descriptors, one can obtain the path to the original Excel file of the PaDEL descriptors with:

print(padel.details)

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

📚 API Documentation

def calculate(mols, show_banner=True, njobs=1, chunksize=None, timeout=None):

Calculates PaDEL molecular descriptors and/or fingerprints. Installs a matching JRE on first use if none is found.

Parameters

• mols : Iterable[Chem.Mol] RDKit molecule objects for which to obtain PaDEL descriptors.
• show_banner : bool Displays default notice about PaDEL descriptors.
• njobs : int Number of concurrent processes used to calculate descriptors in parallel; must not exceed the number of available CPU cores. Each spawned Java process is pinned to a single core (-XX:ActiveProcessorCount=1), since parallelism comes from spawning njobs OS processes rather than from letting each JVM oversubscribe the host's full core count.
• chunksize : int | None Number of molecules processed per worker process. If None (default), molecules are auto-balanced across all njobs workers so every worker gets work. Ignored if njobs is 1.
• timeout : float | None Maximum number of seconds allowed for each descriptor/fingerprint calculation for a single molecule. None (default) waits indefinitely. Guards against 3D descriptors (e.g. WHIM) that can hang on degenerate geometries — the offending molecule gets NaN instead of blocking the whole batch.
• return_type : pd.DataFrame Pandas DataFrame containing PaDEL molecular descriptors and/or fingerprints, one row per molecule.

---

Descriptor(name, is_3D)

Metadata holder for a single PaDEL descriptor family (e.g. ALOGP, WHIM). Instances are pre-created and importable by name from PaDEL_pywrapper.descriptor.

Attributes

• name : str Name of the descriptor as known to PaDEL.
• is_3D : bool Whether the descriptor requires 3D molecular coordinates.
• subcomponents : list[str] Names of the individual output columns making up this descriptor.
• description : pd.DataFrame Human-readable description of each sub-component.

---

Fingerprint(name)

Metadata holder for a single PaDEL fingerprint type. Instances are pre-created and importable by name from PaDEL_pywrapper.descriptor; configurable ones (FP, GraphOnlyFP) can be called with size/searchDepth to override the defaults.

Attributes

• name : str Name of the fingerprint as known to PaDEL.
• size : int | None Number of bits, if configurable for this fingerprint type.
• searchDepth : int | None Search depth, if configurable for this fingerprint type.