Welcome to ODDT’s documentation!



  • Python 2.7+ or 3.4+

  • OpenBabel (2.3.2+) or/and RDKit (2016.03)

  • Numpy (1.8+)

  • Scipy (0.14+)

  • Sklearn (0.18+)

  • joblib (0.8+)

  • pandas (0.17.1+)

  • Skimage (0.10+) (optional, only for surface generation)


All installation methods assume that one of toolkits is installed. For detailed installation procedure visit toolkit’s website (OpenBabel, RDKit)

Most convenient way of installing ODDT is using PIP. All required python modules will be installed automatically, although toolkits, either OpenBabel (pip install openbabel) or RDKit need to be installed manually

pip install oddt

If you want to install cutting edge version (master branch from GitHub) of ODDT also using PIP

pip install git+https://github.com/oddt/oddt.git@master

Finally you can install ODDT straight from the source

wget https://github.com/oddt/oddt/archive/0.5.tar.gz
tar zxvf 0.5.tar.gz
cd oddt-0.5/
python setup.py install

Usage Instructions

You can use any supported toolkit united under common API (for reference see Pybel or Cinfony). All methods and software which based on Pybel/Cinfony should be drop in compatible with ODDT toolkits. In contrast to its predecessors, which were aimed to have minimalistic API, ODDT introduces extended methods and additional handles. This extensions allow to use toolkits at all its grace and some features may be backported from others to introduce missing functionalities. To name a few:

  • coordinates are returned as Numpy Arrays

  • atoms and residues methods of Molecule class are lazy, ie. not returning a list of pointers, rather an object which allows indexing and iterating through atoms/residues

  • Bond object (similar to Atom)

  • atom_dict, ring_dict, res_dict - comprehensive Numpy Arrays containing common information about given entity, particularly useful for high performance computing, ie. interactions, scoring etc.

  • lazy Molecule (asynchronous), which is not converted to an object in reading phase, rather passed as a string and read in when underlying object is called

  • pickling introduced for Pybel Molecule (internally saved to mol2 string)

Atom, residues, bonds iteration

One of the most common operation would be iterating through molecules atoms

mol = oddt.toolkit.readstring('smi', 'c1cccc1')
for atom in mol:


mol.atoms, returns an object (AtomStack) which can be access via indexes or iterated

Iterating over residues is also very convenient, especially for proteins

for res in mol.residues:

Additionally residues can fetch atoms belonging to them:

for res in mol.residues:
    for atom in res:

Bonds are also iterable, similar to residues:

for bond in mol.bonds:
    for atom in bond:

Reading molecules

Reading molecules is mostly identical to Pybel.

Reading from file

for mol in oddt.toolkit.readfile('smi', 'test.smi'):

Reading from string

mol = oddt.toolkit.readstring('smi', 'c1ccccc1 benzene'):


You can force molecules to be read in asynchronously, aka “lazy molecules”. Current default is not to produce lazy molecules due to OpenBabel’s Memory Leaks in OBConverter. Main advantage of lazy molecules is using them in multiprocessing, then conversion is spreaded on all jobs.

Reading molecules from file in asynchronous manner

for mol in oddt.toolkit.readfile('smi', 'test.smi', lazy=True):

This example will execute instantaneously, since no molecules were evaluated.

Numpy Dictionaries - store your molecule as an uniform structure

Most important and handy property of Molecule in ODDT are Numpy dictionaries containing most properties of supplied molecule. Some of them are straightforward, other require some calculation, ie. atom features. Dictionaries are provided for major entities of molecule: atoms, bonds, residues and rings. It was primarily used for interactions calculations, although it is applicable for any other calculation. The main benefit is marvelous Numpy broadcasting and subsetting.

Each dictionary is defined as a format in Numpy.


Atom basic information

  • coords’, type: float32, shape: (3) - atom coordinates

  • charge’, type: float32 - atom’s charge

  • atomicnum’, type: int8 - atomic number

  • atomtype’, type: a4 - Sybyl atom’s type

  • hybridization’, type: int8 - atoms hybrydization

  • neighbors’, type: float32, shape: (4,3) - coordinates of non-H neighbors coordinates for angles (max of 4 neighbors should be enough)

Residue information for current atom

  • resid’, type: int16 - residue ID

  • resnumber’, type: int16 - residue number

  • resname’, type: a3 - Residue name (3 letters)

  • isbackbone’, type: bool - is atom part of backbone

Atom properties

  • isacceptor’, type: bool - is atom H-bond acceptor

  • isdonor’, type: bool - is atom H-bond donor

  • isdonorh’, type: bool - is atom H-bond donor Hydrogen

  • ismetal’, type: bool - is atom a metal

  • ishydrophobe’, type: bool - is atom hydrophobic

  • isaromatic’, type: bool - is atom aromatic

  • isminus’, type: bool - is atom negatively charged/chargable

  • isplus’, type: bool - is atom positively charged/chargable

  • ishalogen’, type: bool - is atom a halogen

Secondary structure

  • isalpha’, type: bool - is atom a part of alpha helix

  • isbeta’, type: bool' - is atom a part of beta strand


  • centroid’, type: float32, shape: 3 - coordinates of ring’s centroid

  • vector’, type: float32, shape: 3 - normal vector for ring

  • isalpha’, type: bool - is ring a part of alpha helix

  • isbeta’, type: bool' - is ring a part of beta strand


  • id’, type: int16 - residue ID

  • resnumber’, type: int16 - residue number

  • resname’, type: a3 - Residue name (3 letters)

  • N’, type: float32, shape: 3 - cordinates of backbone N atom

  • CA’, type: float32, shape: 3 - cordinates of backbone CA atom

  • C’, type: float32, shape: 3 - cordinates of backbone C atom

  • isalpha’, type: bool - is residue a part of alpha helix

  • isbeta’, type: bool' - is residue a part of beta strand


All aforementioned dictionaries are generated “on demand”, and are cached for molecule, thus can be shared between calculations. Caching of dictionaries brings incredible performance gain, since in some applications their generation is the major time consuming task.

Get all acceptor atoms:


Interaction Fingerprints

Module, where interactions between two molecules are calculated and stored in fingerprint.

The most common usage

Firstly, loading files

protein = next(oddt.toolkit.readfile('pdb', 'protein.pdb'))
protein.protein = True
ligand = next(oddt.toolkit.readfile('sdf', 'ligand.sdf'))


You have to mark a variable with file as protein, otherwise You won’t be able to get access to e.g. ‘resname; , ‘resid’ etc. It can be done as above.

File with more than one molecule

mols = list(oddt.toolkit.readfile('sdf', 'ligands.sdf'))

When files are loaded, You can check interactions between molecules. Let’s find out, which amino acids creates hydrogen bonds

protein_atoms, ligand_atoms, strict = hbonds(protein, ligand)

Or check hydrophobic contacts between molecules

protein_atoms, ligand_atoms = hydrophobic_contacts(protein, ligand)
print(protein_atoms, ligand_atoms)

But instead of checking interactions one by one, You can use fingerprints module.

IFP = InteractionFingerprint(ligand, protein)
SIFP = SimpleInteractionFingerprint(ligand, protein)

Very often we’re looking for similar molecules. We can easily accomplish this by e.g.

results = []
reference = SimpleInteractionFingerprint(ligand, protein)
for el in query:
    fp_query = SimpleInteractionFingerprint(el, protein)
    # similarity score for current query
    cur_score = dice(reference, fp_query)
    # score is the lowest, required similarity
    if cur_score > score:
return results

Molecular shape comparison

Three methods for molecular shape comparison are supported: USR and its two derivatives: USRCAT and Electroshape.

  • USR (Ultrafast Shape Recognition) - function usr(molecule)

    Ballester PJ, Richards WG (2007). Ultrafast shape recognition to search compound databases for similar molecular shapes. Journal of computational chemistry, 28(10):1711-23. http://dx.doi.org/10.1002/jcc.20681

  • USRCAT (USR with Credo Atom Types) - function usr_cat(molecule)

    Adrian M Schreyer, Tom Blundell (2012). USRCAT: real-time ultrafast shape recognition with pharmacophoric constraints. Journal of Cheminformatics, 2012 4:27. http://dx.doi.org/10.1186/1758-2946-4-27

  • Electroshape - function electroshape(molecule)

    Armstrong, M. S. et al. ElectroShape: fast molecular similarity calculations incorporating shape, chirality and electrostatics. J Comput Aided Mol Des 24, 789-801 (2010). http://dx.doi.org/doi:10.1007/s10822-010-9374-0

    Aside from spatial coordinates, atoms’ charges are also used as the fourth dimension to describe shape of the molecule.

To find most similar molecules from the given set, each of these methods can be used.

Loading files:

query = next(oddt.toolkit.readfile('sdf', 'query.sdf'))
database = list(oddt.toolkit.readfile('sdf', 'database.sdf'))

Example code to find similar molecules:

results = []
query_shape = usr(query)
for mol in database:
    mol_shape = usr(mol)
    similarity = usr_similarity(query_shape, mol_shape)
    if similarity > 0.7:

To use another method, replace usr(mol) with usr_cat(mol) or electroshape(mol).

ODDT command line interface (CLI)

There is an oddt command to interface with Open Drug Discovery Toolkit from terminal, without any programming knowleadge. It simply reproduces oddt.virtualscreening.virtualscreening. One can filter, dock and score ligands using methods implemented or compatible with ODDT. All positional arguments are treated as input ligands, whereas output must be assigned using -O option (following obabel convention). Input and output formats are defined using -i and -o accordingly. If output format is present and no output file is assigned, then molecules are printed to STDOUT.

To list all the available options issue -h option:

oddt_cli -h
  1. Docking ligand using Autodock Vina (construct box using ligand from crystal structure) with additional RFscore v2 rescoring:

oddt_cli input_ligands.sdf --dock autodock_vina --receptor rec.mol2 --auto_ligand crystal_ligand.mol2 --score rfscore_v2 -O output_ligands.sdf
  1. Filtering ligands using Lipinski RO5 and PAINS. Afterwards dock with Autodock Vina:

oddt_cli input_ligands.sdf --filter ro5 --filter pains --dock autodock_vina --receptor rec.mol2 --auto_ligand crystal_ligand.mol2 -O output_ligands.sdf
  1. Dock with Autodock Vina, with precise box position and dimensions. Fix seed for reproducibility and increase exhaustiveness:

oddt_cli ampc/actives_final.mol2.gz --dock autodock_vina --receptor ampc/receptor.pdb --size '(8,8,8)' --center '(1,2,0.5)' --exhaustiveness 20 --seed 1 -O ampc_docked.sdf
  1. Rescore ligands using 3 versions of RFscore and pre-trained scoring function (either pickle from ODDT or any other SF implementing oddt.scoring.scorer API):

oddt_cli docked_ligands.sdf --receptor rec.mol2 --score rfscore_v1 --score rfscore_v2 --score rfscore_v3 --score TrainedNN.pickle -O docked_ligands_rescored.sdf

Development and contributions guide

1. Indicies All indicies within toolkit are 0-based, but for backward compatibility with OpenBabel there is mol.idx property. If you develop using ODDT you are encouraged to use 0-based indicies and/or mol.idx0 and mol.idx1 properties to be exact which convention you adhere to. Otherwise you can run into bags which are hard to catch, when writing toolkit independent code.


To be announced.