pysmiles: The lightweight and pure-python SMILES reader and writer

This is a small project I started because I couldn't find any SMILES reader or writer that was easy to install (read: Python only). Currently, the writer is extremely basic, and although it should produce valid SMILES they won't be pretty, but see also issue #17. The reader is in a better state, and should be usable.

SMILES strings are assumed to be as specified by the OpenSmiles standard.

Molecules

Molecules are depicted as Networkx graphs. Atoms are the nodes of the graph, and bonds are the edges. Nodes can have the following attributes:

element: str. This describes the element of the atom. Defaults to '*' meaning unknown.
aromatic: bool. Whether the atom is part of an (anti)-aromatic system. Defaults to False.
isotope: float. The mass of the atom. Defaults to unknown.
hcount: int. The number of implicit hydrogens attached to this atom. Defaults to 0.
charge: int. The charge of this atom. Defaults to 0.
class: int. The "class" of this atom. Defaults to 0.
rs_isomer and ez_isomer: See below.
stereo: Not yet used, but reserved for more interpretable stereochemical information

Edges have the following attributes:

order: Number. The bond order. 1.5 is used for aromatic bonds. Defaults to 1.

Reading SMILES

The function read_smiles(smiles, explicit_hydrogen=False, zero_order_bonds=True, reinterpret_aromatic=True, strict=True) can be used to parse a SMILES string. It should not be used to validate whether a string is a valid SMILES string --- the function does very little validation whether your SMILES string makes chemical sense. Edges in the created molecule will always have an 'order' attribute. Nodes will have the relevant attributes in so far they are specified. Atoms for which the element is not known (*) will not have an element attribute.

explicit_hydrogen determines whether hydrogen atoms should be represented as explicit nodes in the created molecule, or implicit in the 'hcount' attribute.
zero_order_bonds determines whether zero-order bonds (.) in the SMILES string should result in edges in the produced molecule.
reinterpret_aromatic determines whether aromaticity should be reinterpreted, and determined from the constructed molecule, or whether the aromaticity specifications from the SMILES string (lower case elements) should be taken as leading. If True, will also set bond orders to 1 for bonds that are not part of an aromatic ring and have a bond order of 1.5. See also below. If False, will create a molecule using only the information in the SMILES string.
strict determines whether the function be a bit more critical about what constitutes a valid SMILES string. In particular, checks whether the specified aromatic fragments can be kekulized, and whether atom valency is sensible.

Stereochemical information

Tetrahedral chirality is stored on nodes in the 'rs_isomer' attribute. It consists of 4 node indices. The first, together with the central atom, indicates the axis of rotation/observation. The last three indices indicate the direction of rotation in counter-clockwise order. For example, when parsing the SMILES N[C@](Br)(O)C node 1 (the central carbon) will have 'rs_isomer' attribute (0, 2, 3, 4). This means when looking along the axis (0, 1) (nitrogen to carbon), the nodes 2 (bromine), 3 (oxygen), 4 (methyl) will be in counter-clockwise order. If the central atom has an implicit hydrogen (which of course does not have its own node index) we use the index of the central atom instead. There is currently no way to easily convert this to R/S labels.

E/Z chirality around double bonds is encoded on nodes in the 'ez_isomer' attribute. This attribute consists of 4 node indices and 1 string indication either 'cis' or 'trans'. The node indices indicate the dihedral angle. For example, when parsing the SMILES Br/C=C\F node 0 (the bromine) will have 'ez_isomer' attribute [0, 1, 2, 3, 'cis']. This means the dihedral angle of the axes (0, 1) and (2, 3) around the central axis (1, 2) is 0 degrees. There is currently no way to easily convert this to E/Z labels.

Aromaticity

Aromaticity in SMILES has little to do with how an organic chemist might interpret the term. Here we use "aromatic" to mean "delocalization induced molecular equivalence" (DIME). DIME means that molecular fragments that are equivalent under delocalization should get the same representation. As an example, C1=CC=CC=C1 and C=1C=CC=CC1 are equivalent, and should both be represented as c1ccccc1. On the other hand, pyrolle (c1c[nH]cc1) can only be represented as C1=CNC=C1, and is not considered aromatic. A series of blogposts with a decent writeup on the subject can be found here.

Despite how simple it seems at a glance, here be dragons lurking in the details. The main problem is that in order to determine whether an edge is aromatic, you need to generate all simple cycles. To illustrate why this is a problem, let's look at buckminsterfullerene: it contains a staggering 15,024,073 cycles! It's really not feasible to investigate all of those separately.

Therefor we approximate the aromaticity for ring systems that are too large. By default we consider 30 atoms large. For the exact algorithm, please see the docstring of dekekulize. The approximation usually works fine, but may erroneously mark some triangle edges as aromatic. If you have a SMILES string with a large ring system for which the approximation does not work, please open an issue. As a workaround, you can read the SMILES with reinterpret_aromatic=False, and manually call correct_aromatic_rings with more appropriate thresholds.

A second complication comes from wildcard (*) atoms. If possible we try to kekulize the provided molecule without using them, but there may be edge cases where this does not work as intended.

Writing SMILES

The function write_smiles(molecule, default_element='*', start=None) can be used to write SMILES strings from a molecule. The function does not check whether your molecule makes chemical sense. Instead, it writes a SMILES representation of the molecule you provided, and nothing else. It does not yet know how to write stereochemical information that is present in your molecule, and will log a warning if this happens. See below if you need to silence these.

default_element is the element to use for nodes that do not have an 'element' attribute.
start is the key of the node where the depth first traversal should be started. Something clever is done if not specified.

Additional functions

In addition to these two core functions, four more functions are exposed that can help in creating chemically relevant molecules with minimal work.

fill_valence(mol, respect_hcount=True, respect_bond_order=True, max_bond_order=3) This function will fill the valence of all atoms in your molecule by incrementing the 'hcount' and, if specified, bond orders.
- repect_hcount: bool. Whether existing hcounts can be overwritten.
- respect_bond_order: bool. Whether bond orders can be changed.
- max_bond_order: int. The maximum bond order that will be set.
add_explicit_hydrogens(mol) This function transforms implicit hydrogens, specified by 'hcount' attributes, to explicit nodes.
remove_explicit_hydrogens(mol) This function does the inverse of add_explicit_hydrogens: it will remove explicit hydrogen nodes and add them to the relevant 'hcount' attributes.
correct_aromatic_rings(mol, strict=True, estimation_threshold=None, max_ring_size=None) This function marks all (anti)-aromatic atoms in your molecule, and sets all bonds between (anti)-aromatic atoms to order 1.5.

It works by first finding all atoms that could participate in an aromatic system. These are atoms that can make a double bond without introducing a charge. Then, it tries to assign alternating double and single bonds
between the atoms found. If it cannot and strict is True an error will be raised. Otherwise, it starts to look for cycles that consist of such alternating double and single bonds. Atoms in these cycles will be marked as aromatic, and these edges will get a bond order of 1.5. The relevant atoms that are not in any cycle will be marked as not-aromatic, and these bonds will be assigned alternating bond orders of 1 and 2.
kekulize(mol) Assigns alternating single and double bonds to all aromatic regions in mol. Will raise an error if it cannot.
dekekulize(mol, estimation_threshold=None, max_ring_size=None) Finds all cycles in mol that consist of alternating single and double bonds, and marks them as aromatic.

Examples

Reading

from pysmiles import read_smiles

smiles = 'C1CC[13CH2]CC1C1CCCCC1'
mol = read_smiles(smiles)

print(mol.nodes(data='element'))
# [(0, 'C'),
#  (1, 'C'),
#  (2, 'C'),
#  (3, 'C'),
#  (4, 'C'),
#  (5, 'C'),
#  (6, 'C'),
#  (7, 'C'),
#  (8, 'C'),
#  (9, 'C'),
#  (10, 'C'),
#  (11, 'C')]
print(mol.nodes(data='hcount'))
# [(0, 2),
#  (1, 2),
#  (2, 2),
#  (3, 2),
#  (4, 2),
#  (5, 1),
#  (6, 1),
#  (7, 2),
#  (8, 2),
#  (9, 2),
#  (10, 2),
#  (11, 2)]

mol_with_H = read_smiles(smiles, explicit_hydrogen=True)
print(mol_with_H.nodes(data='element'))
# [(0, 'C'),
#  (1, 'C'),
#  (2, 'C'),
#  (3, 'C'),
#  (4, 'C'),
#  (5, 'C'),
#  (6, 'C'),
#  (7, 'C'),
#  (8, 'C'),
#  (9, 'C'),
#  (10, 'C'),
#  (11, 'C'),
#  (12, 'H'),
#  (13, 'H'),
#  (14, 'H'),
#  (15, 'H'),
#  (16, 'H'),
#  (17, 'H'),
#  (18, 'H'),
#  (19, 'H'),
#  (20, 'H'),
#  (21, 'H'),
#  (22, 'H'),
#  (23, 'H'),
#  (24, 'H'),
#  (25, 'H'),
#  (26, 'H'),
#  (27, 'H'),
#  (28, 'H'),
#  (29, 'H'),
#  (30, 'H'),
#  (31, 'H'),
#  (32, 'H'),
# (33, 'H')]

Writing

import networkx as nx
from pysmiles import write_smiles, fill_valence

mol = nx.Graph()
mol.add_edges_from([(0, 1), (1, 2), (1, 3), (3, 4), (1, 5), (3, 6)])
for idx, ele in enumerate('CCCCOCO'):
    mol.nodes[idx]['element'] = ele
mol.nodes[4]['charge'] = -1
mol.nodes[4]['hcount'] = 0
mol.edges[3, 6]['order'] = 2

print(write_smiles(mol))
# [O-]C(=O)C([C])([C])[C]
fill_valence(mol, respect_hcount=True)
print(write_smiles(mol))
# [O-]C(=O)C(C)(C)C

Logging

Pysmiles uses the python logging module to log warnings. If you need to silence these, you can use the following snippet in your code:

import logging
logging.getLogger('pysmiles').setLevel(logging.CRITICAL)  # Anything higher than warning

Limitations

The writer produces non-recommended SMILES strings (as per OpenSmiles).
The writer is better described as a "serializer": if the graph provided doesn't make chemical sense the produced "SMILES" string will be an exact representation of that graph. Because of this, the SMILES string will be invalid though.
The writer cannot deal with stereochemistry, and ignores it.
There is no way to easily transform stereochemical information to human-readable labels like R/S/E/Z.
Extended stereochemical centers, such as described by NC(Br)=[C@]=C(O)C are not supported.
Although the SMILES F/C(/Br)=C\F is valid, pysmiles does not understand it.
It only processes SMILES. This might later be extended to e.g. InChi, SLN, SMARTS, etc.

Requirements

networkx

Similar projects

There are more python projects that deal with SMILES, and I try to list at least some of them here. If yours is missing, feel free to open up a PR.

PySMILE: A similar named project, capable of encoding/decoding SMILE format objects. Doesn't deal with SMILES.
RDKit: A collection of cheminformatics and machine-learning software, capable of reading and writing SMILES, InChi, and others.
OpenEye Chem toolkit: The OpenEye chemistry toolkit is a programming library for chemistry and cheminformatics. It is capable of dealing with (canonical) SMILES and InChi.

License

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

    http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.

pckroon / pysmiles

readme