The RDKit Book

Misc Cheminformatics Topics

Aromaticity

Aromaticity is one of those unpleasant topics that is simultaneously simple and impossibly complicated. Since neither experimental nor theoretical chemists can agree with each other about a definition, it’s necessary to pick something arbitrary and stick to it. This is the approach taken in the RDKit.

Instead of using patterns to match known aromatic systems, the aromaticity perception code in the RDKit uses a set of rules. The rules are relatively straightforward.

Aromaticity is a property of atoms and bonds in rings. An aromatic bond must be between aromatic atoms, but a bond between aromatic atoms does not need to be aromatic.

For example the fusing bonds here are not considered to be aromatic by the RDKit:

_images/picture_9.png
>>> from rdkit import Chem
>>> m = Chem.MolFromSmiles('C1=CC2=C(C=C1)C1=CC=CC=C21')
>>> m.GetAtomWithIdx(3).GetIsAromatic()
True
>>> m.GetAtomWithIdx(6).GetIsAromatic()
True
>>> m.GetBondBetweenAtoms(3,6).GetIsAromatic()
False

The RDKit supports a number of different aromaticity models and allows the user to define their own by providing a function that assigns aromaticity.

The RDKit Aromaticity Model

A ring, or fused ring system, is considered to be aromatic if it obeys the 4N+2 rule. Contributions to the electron count are determined by atom type and environment. Some examples:

Fragment

Number of pi electrons

c(a)a

1

n(a)a

1

An(a)a

2

o(a)a

2

s(a)a

2

se(a)a

2

te(a)a

2

O=c(a)a

0

N=c(a)a

0

*(a)a

0, 1, or 2

Notation a: any aromatic atom; A: any atom, include H; *: a dummy atom

Notice that exocyclic bonds to electronegative atoms “steal” the valence electron from the ring atom and that dummy atoms contribute whatever count is necessary to make the ring aromatic.

The use of fused rings for aromaticity can lead to situations where individual rings are not aromatic, but the fused system is. An example of this is azulene:

_images/picture_8.png

An extreme example, demonstrating both fused rings and the influence of exocyclic double bonds:

_images/picture_7.png
>>> m=Chem.MolFromSmiles('O=C1C=CC(=O)C2=C1OC=CO2')
>>> m.GetAtomWithIdx(6).GetIsAromatic()
True
>>> m.GetAtomWithIdx(7).GetIsAromatic()
True
>>> m.GetBondBetweenAtoms(6,7).GetIsAromatic()
False

A special case, heteroatoms with radicals are not considered candidates for aromaticity:

_images/picture_10.png
>>> m = Chem.MolFromSmiles('C1=C[N]C=C1')
>>> m.GetAtomWithIdx(0).GetIsAromatic()
False
>>> m.GetAtomWithIdx(2).GetIsAromatic()
False
>>> m.GetAtomWithIdx(2).GetNumRadicalElectrons()
1

Charged carbons with radicals are also not considered:

_images/picture_12.png
>>> m = Chem.MolFromSmiles('C1=CC=CC=C[C+]1')
>>> m.GetAtomWithIdx(0).GetIsAromatic()
False
>>> m.GetAtomWithIdx(6).GetIsAromatic()
False
>>> m.GetAtomWithIdx(6).GetFormalCharge()
1
>>> m.GetAtomWithIdx(6).GetNumRadicalElectrons()
1

Neutral carbons with radicals, however, are still considered:

_images/picture_11.png
>>> m = Chem.MolFromSmiles('C1=[C]NC=C1')
>>> m.GetAtomWithIdx(0).GetIsAromatic()
True
>>> m.GetAtomWithIdx(1).GetIsAromatic()
True
>>> m.GetAtomWithIdx(1).GetNumRadicalElectrons()
1

The Simple Aromaticity Model

This one is quite simple: only five- and six-membered simple rings are considered candidates for aromaticity. The same electron-contribution counts listed above are used.

The MDL Aromaticity Model

This isn’t well documented (at least not publicly), so we tried to reproduce what’s provided in the oechem documentation (https://docs.eyesopen.com/toolkits/python/oechemtk/aromaticity.html)

  • fused rings (i.e. azulene) can be aromatic

  • five-membered rings are not aromatic (though they can be part of fused aromatic systems)

  • only C and N can be aromatic

  • only one electron donors are accepted

  • atoms with exocyclic double bonds are not aromatic

Note: For reasons of computational expediency, aromaticity perception is only done for fused-ring systems where all members are at most 24 atoms in size.

SMILES Support and Extensions

The RDKit covers all of the standard features of Daylight SMILES [2] as well as some useful extensions.

Here’s the (likely partial) list of extensions:

Aromaticity

te (aromatic Te) is accepted. Here is an example with tellurophene-2-carboxylic acid:

>>> m = Chem.MolFromSmiles('OC(=O)c1[te]ccc1')
>>> m.GetAtomWithIdx(4).GetIsAromatic()
True

Dative bonds

<- and -> create a dative bond between the atoms, direction does matter.

Here’s an example of a bipy-copper complex:

>>> bipycu = Chem.MolFromSmiles('c1cccn->2c1-c1n->3cccc1.[Cu]23(Cl)Cl')
>>> bipycu.GetBondBetweenAtoms(4,12).GetBondType()
rdkit.Chem.rdchem.BondType.DATIVE
>>> Chem.MolToSmiles(bipycu)
'Cl[Cu]1(Cl)<-n2ccccc2-c2ccccn->12'

Dative bonds have the special characteristic that they don’t affect the valence on the start atom, but do affect the end atom. So in this case, the N atoms involved in the dative bond have the valence of 3 that we expect from bipy, while the Cu has a valence of 4:

>>> bipycu.GetAtomWithIdx(4).GetTotalValence()
3
>>> bipycu.GetAtomWithIdx(12).GetTotalValence()
4

Ring closures

%(N) notation is supported for ring closures, where N is a single digit %(N) up to five digits %(NNNNN). Here is an example:

>>> m = Chem.MolFromSmiles('C%(1000)OC%(1000)')
>>> m.GetAtomWithIdx(0).IsInRing()
True
>>> m.GetAtomWithIdx(2).IsInRing()
True

Specifying atoms by atomic number

The [#6] construct from SMARTS is supported in SMILES.

Quadruple bonds

The token $ can be used to represent quadruple bonds in SMILES and SMARTS.

CXSMILES/CXSMARTS extensions

The RDKit supports parsing and writing a subset of the extended SMILES/SMARTS functionality introduced by ChemAxon [4].

The features which are parsed include:

  • atomic coordinates ()

  • atomic values $_AV:

  • atomic labels/aliases $ (recognized aliases are _AP, star_e, Q_e, QH_p, AH_P, X_p, XH_p, M_p, MH_p, *)

  • atomic properties atomprop

  • coordinate/dative bonds C (these are translated into dative bonds)

  • radicals ^

  • enhanced stereo (these are converted into StereoGroups)

  • linknodes LN

  • variable/multi-center attachments m

  • ring bond count specifications rb

  • non-hydrogen substitution count specifications s

  • unsaturation specification u

  • wedged bonds (only when atomic coordinates are present): wU, wD

  • wiggly bonds w

  • double bond stereo (only for ring bonds) c, t, ctu

  • SGroup Data SgD

  • polymer SGroups Sg

  • SGroup Hierarchy SgH

The features which are written by rdkit.Chem.rdmolfiles.MolToCXSmiles() and rdkit.Chem.rdmolfiles.MolToCXSmarts() (note the specialized writer functions) include:

  • atomic coordinates

  • atomic values

  • atomic labels

  • atomic properties

  • dative bonds (only if dative bonds are not also being written to the SMILES/SMARTS)

  • radicals

  • enhanced stereo

  • linknodes

  • wedged bonds (only when atomic coordinates are also written)

  • wiggly bonds

  • double bond stereo (only for ring bonds)

  • SGroup Data

  • polymer SGroups

  • SGroup Hierarchy

>>> m = Chem.MolFromSmiles('OC')
>>> m.GetAtomWithIdx(0).SetProp('p1','2')
>>> m.GetAtomWithIdx(1).SetProp('p1','5')
>>> m.GetAtomWithIdx(1).SetProp('p2','A1')
>>> m.GetAtomWithIdx(0).SetProp('atomLabel','O1')
>>> m.GetAtomWithIdx(1).SetProp('atomLabel','C2')
>>> Chem.MolToCXSmiles(m)
'CO |$C2;O1$,atomProp:0.p1.5:0.p2.A1:1.p1.2|'

Reading molecule names

If the SMILES/SMARTS and the optional CXSMILES extensions are followed by whitespace and another string, the SMILES/SMARTS parsers will interpret this as the molecule name:

>>> m = Chem.MolFromSmiles('CO carbon monoxide')
>>> m.GetProp('_Name')
'carbon monoxide'
>>> m2 = Chem.MolFromSmiles('CO |$C2;O1$| carbon monoxide')
>>> m2.GetAtomWithIdx(0).GetProp('atomLabel')
'C2'
>>> m2.GetProp('_Name')
'carbon monoxide'

This can be disabled while still parsing the CXSMILES:

>>> ps = Chem.SmilesParserParams()
>>> ps.parseName = False
>>> m3 = Chem.MolFromSmiles('CO |$C2;O1$| carbon monoxide',ps)
>>> m3.HasProp('_Name')
0
>>> m3.GetAtomWithIdx(0).GetProp('atomLabel')
'C2'

Note that if you disable CXSMILES parsing but pass in a string which includes CXSMILES it will be interpreted as (part of) the name:

>>> ps = Chem.SmilesParserParams()
>>> ps.allowCXSMILES = False
>>> m4 = Chem.MolFromSmiles('CO |$C2;O1$| carbon monoxide',ps)
>>> m4.GetProp('_Name')
'|$C2;O1$| carbon monoxide'

Finally, if you disable parsing of both CXSMILES and names, then extra text in the SMILES/SMARTS string will result in errors: .. doctest:

>>> ps = Chem.SmilesParserParams()
>>> ps.allowCXSMILES = False
>>> ps.parseName = False
>>> m5 = Chem.MolFromSmiles('CO |$C2;O1$| carbon monoxide',ps)
>>> m5 is None
True
>>> m5 = Chem.MolFromSmiles('CO carbon monoxide',ps)
>>> m5 is None
True

The examples in this sectin all used the SMILES parser, but the SMARTS parser behaves the same way.

SMARTS Support and Extensions

The RDKit covers most of the standard features of Daylight SMARTS [3] as well as some useful extensions.

Here’s the (hopefully complete) list of SMARTS features that are not supported:

  • Non-tetrahedral chiral classes

  • the @? operator

  • explicit atomic masses (though isotope queries are supported)

  • component level grouping requiring matches in different components, i.e. (C).(C)

Here’s the (likely partial) list of extensions:

Hybridization queries

  • ^0 matches S hybridized atoms

  • ^1 matches SP hybridized atoms

  • ^2 matches SP2 hybridized atoms

  • ^3 matches SP3 hybridized atoms

  • ^4 matches SP3D hybridized atoms

  • ^5 matches SP3D2 hybridized atoms

>> Chem.MolFromSmiles('CC=CF').GetSubstructMatches(Chem.MolFromSmarts('[^2]'))
((1,), (2,))

Dative bonds

<- and -> match the corresponding dative bonds, direction does matter.

>>> Chem.MolFromSmiles('C1=CC=CC=N1->[Fe]').GetSubstructMatches(Chem.MolFromSmarts('[#7]->*'))
((5, 6),)
>>> Chem.MolFromSmiles('C1=CC=CC=N1->[Fe]').GetSubstructMatches(Chem.MolFromSmarts('*<-[#7]'))
((6, 5),)

Heteroatom neighbor queries

  • the atom query z matches atoms that have the specified number of heteroatom (i.e. not C or H) neighbors. For example, z2 would match the second C in CC(=O)O.

  • the atom query Z matches atoms that have the specified number of aliphatic heteroatom (i.e. not C or H) neighbors.

>>> Chem.MolFromSmiles('O=C(O)c1nc(O)ccn1').GetSubstructMatches(Chem.MolFromSmarts('[z2]'))
((1,), (3,), (5,))
>>> Chem.MolFromSmiles('O=C(O)c1nc(O)ccn1').GetSubstructMatches(Chem.MolFromSmarts('[Z2]'))
((1,),)
>>> Chem.MolFromSmiles('O=C(O)c1nc(O)ccn1').GetSubstructMatches(Chem.MolFromSmarts('[Z1]'))
((5,),)

Range queries

Ranges of values can be provided for many query types that expect numeric values. The query types that currently support range queries are: D, h, r, R, v, x, X, z, Z, +, -

Here are some examples:
  • D{2-4} matches atoms that have between 2 and 4 (inclusive) explicit connections.

  • D{-3} matches atoms that have less than or equal to 3 explicit connections.

  • D{2-} matches atoms that have at least 2 explicit connections.

>>> Chem.MolFromSmiles('CC(=O)OC').GetSubstructMatches(Chem.MolFromSmarts('[z{1-}]'))
((1,), (4,))
>>> Chem.MolFromSmiles('CC(=O)OC').GetSubstructMatches(Chem.MolFromSmarts('[D{2-3}]'))
((1,), (3,))
>>> Chem.MolFromSmiles('CC(=O)OC.C').GetSubstructMatches(Chem.MolFromSmarts('[D{-2}]'))
((0,), (2,), (3,), (4,), (5,))

SMARTS Reference

Note that the text versions of the tables below include some backslash characters to escape special characters. This is a wart from the documentation system we are using. Please ignore those characters.

Atoms

Primitive

Property

“Default value”

Range?

Notes

a

“aromatic atom”

A

“aliphatic atom”

d

“non-hydrogen degree”

1

Y

extension

D

“explicit degree”

1

Y

h

“number of implicit hs”

>0

Y

H

“total number of Hs”

1

r

“size of smallest SSSR ring”

>0

Y

R

“number of SSSR rings”

>0

Y

v

“total valence”

1

Y

x

“number of ring bonds”

>0

Y

X

“total degree”

1

Y

z

“number of heteroatom neighbors”

>0

Y

extension

Z

“number of aliphatic heteroatom neighbors”

>0

Y

extension

*

“any atom”

+

“positive charge”

1

Y

++

“+2 charge”

-

“negative charge”

1

Y

--

“-2 charge”

^0

“S hybridized”

n/a

N

extension

^1

“SP hybridized”

n/a

N

extension

^2

“SP2 hybridized”

n/a

N

extension

^3

“SP3 hybridized”

n/a

N

extension

^4

“SP3D hybridized”

n/a

N

extension

^5

“SP3D2 hybridized”

n/a

N

extension

Bonds

Primitive

Property

Notes

“”

“single or aromatic”

“unspecified bonds”

-

single

=

double

#

triple

:

aromatic

~

“any bond”

@

“ring bond”

/

“directional”

\

“directional”

->

“dative right”

extension

<-

“dative left”

extension

Mol/SDF Support and Extensions

The RDKit covers an extensive subset of the features in the V2000 and V3000 CTAB specfication. This subset should be better documented.

Here are the non-element atom queries that are supported:
  • A: any heavy atom

  • Q: any non-carbon heavy atom

  • *: unspecfied (interpreted as any atom)

  • L: (V2000): atom list

  • AH: (ChemAxon Extension) any atom

  • QH: (ChemAxon Extension) any non-carbon atom

  • X: (ChemAxon Extension) halogen

  • XH: (ChemAxon Extension) halogen or hydrogen

  • M: (ChemAxon Extension) metal (“contains alkali metals, alkaline earth metals, transition

    metals, actinides, lanthanides, poor(basic) metals, Ge, Sb, and Po”)

  • MH: (ChemAxon Extension) metal or hydrogen

Here’s a partial list of the features that are supported:
  • enhanced stereochemistry (V3000 only)

  • Sgroups: Sgroups are read and written, but interpretation of their contents is still very much a work in progress

  • Dative bonds in V2000 (type 9), despite them not being part of the standard, we support them because they frequently show up in real-world data

Ring Finding and SSSR

[Section taken from “Getting Started” document]

As others have ranted about with more energy and eloquence than I intend to, the definition of a molecule’s smallest set of smallest rings is not unique. In some high symmetry molecules, a “true” SSSR will give results that are unappealing. For example, the SSSR for cubane only contains 5 rings, even though there are “obviously” 6. This problem can be fixed by implementing a small (instead of smallest) set of smallest rings algorithm that returns symmetric results. This is the approach that we took with the RDKit.

Because it is sometimes useful to be able to count how many SSSR rings are present in the molecule, there is a GetSSSR function, but this only returns the SSSR count, not the potentially non-unique set of rings.

For situations where you just care about knowing whether or not atoms/bonds are in rings, the RDKit provides the function rdkit.Chem.rdmolops.FastFindRings(). This does a depth-first traversal of the molecule graph and identifies atoms and bonds that are in rings.

Stereochemistry

Types of stereochemistry supported

The RDKit currently fully supports tetrahedral atomic stereochemistry and cis/trans stereochemistry at double bonds. There is partial support for non-tetrahedral stereochemistry, see the section Support for non-tetrahedral atomic stereochemistry.

Identification of potential stereoatoms/stereobonds

As of the 2020.09 release the RDKit has two different ways of identifying potential stereoatoms/stereobonds:

  1. The legacy approach: AssignStereochemistry(). This approach does a reasonable job of recognizing potential stereocenters, including some para-stereochemistry. It also has the side effect of assigning approximate CIP labels to the atoms/bonds (see below). This is currently the default algorithm.

  2. The new approach: FindPotentialStereo(). The new approach is both more accurate (particularly for para-stereochemistry) and faster. It will become the default in a future RDKit version.

A concrete example of the accuracy improvements arising from the new algorithm:

parastereo2

parastereo1

Both algorithms recognize that the central carbon is a potential stereocenter in the molecule on the left, but the old algorithm is unable to recognize it as a potential stereocenter in the molecule on the right.

Assignment of absolute stereochemistry

As of the 2020.09 release the RDKit has two different ways of assigning absolute stereochemistry labels (CIP labels):

  1. The legacy approach uses an adaptation of an approximate algorithm for assigning CIP codes published by Paul Labute, [12]. The algorithm is reliable for determining whether or not a particular specified stereoatom/stereobond actually is a stereoatom/stereobond, but the CIP codes which it assigns are only truly correct for simple examples. As of the 2020.09 release this is the default algorithm, but this will be changed in a future RDKit release.

  2. The new approach uses an implementation of a much more accurate algorithm, [13]. The new algorithm is more computationally expensive than the old one and does not provide CIP rankings of atoms (the concept of a global ranking of atoms isn’t well defined within the context of the true CIP algorithm). If you’re interested in having a chirality-sensitive ranking of all atoms, you can use the canonical atom ranking code instead.

Stereogenic atoms/bonds

The definitions of potential stereogenic atoms or bonds is inspired by the InChI definitions.

Stereogenic bonds

A double bond is potentially stereogenic if both atoms have at least two heavy atom neighbors and it’s not present in a ring with less than eight atoms.

For example, both of these double bonds are candidates for stereochemistry:

psdb1

psdb2

But this one is not:

_images/potential_stereo_double_bond3.png
Tetrahedral Stereogenic atoms

The following atom types are potential tetrahedral stereogenic atoms:

  • atoms with degree 4

  • atoms with degree 3 and one implicit H

  • P or As with degree 3 or 4

  • An SP3 hybridized N with degree 3 that is not involved in any conjugated bonds and that is in a ring of size 3 or that is shared between at least 3 rings (this last condition is an extension to the InChI rules).

  • S or Se with degree 3 and a total valence of 4 or a total valence of 3 and a net charge of +1.

Brief description of the findPotentialStereo() algorithm

  1. Identify all potential stereogenic atoms and bonds in the molecule. If there aren’t any we don’t need to do anything else.

  2. Foreach potential stereogenic atom: save the original chiral tag and then set the chiral tag to CW. Assign an atom symbol that makes this atom unique from all others (this will be used below in the canonicalization algorithm)

  3. Foreach potential stereogenic bond: assign a bond symbol that makes this bond unique from all others (this will be used below in the canonicalization algorithm)

  4. Determine the canonical atom ranking taking chirality into account, but not breaking ties. This uses the same canonicalization algorithm that’s used to generate SMILES. [14]

  5. Remove the chiral tag from any potential stereogenic atom which has two identically ranked neighbors and set its symbol to the default for that atom

  6. Set the symbol of any double bond which has two identically ranked atoms attached to either end [15] to the default for that bond

  7. If steps 5 and 6 modfied any atoms or bonds, loop back to step 4.

  8. Add any potential stereogenic atom which does not have to identically ranked neighbors to the results

  9. Add any potential stereogenic atom which does not have two identically ranked atoms attached to either end [15] to the results

  10. Return the results

Sources of information about stereochemistry

From SMILES

Atomic stereochemistry can be specified using @, @@, @SP, etc. Potential stereocenters with no information provided are ChiralType::CHI_UNSPECIFIED.

Double-bond stereochemistry is specfied using / and \ to indicate the directionality of the neighboring single bonds. Double bonds with no stereo information provided are BondStereo::STEREONONE.

From Mol

Atomic stereochemistry can be specified using wedged bonds if 2D coordinates are present. If 3D coordinates are present, they are used to set the stereochemistry for stereogenic atoms. Wiggly bonds (CFG=2 in V3000 mol blocks) set the chiral tag of stereogenic start atom to ChiralType::CHI_UNSPECIFIED.

Double-bond stereochemistry is automatically set using the atomic coordinates; this is true for both 2D and 3D coordinates. If a stereogenic double bound is crossed (CFG=2 in V3000 mol blocks) or has an adjacent wiggly single bond (CFG=2 in V3000 mol blocks), then it will be BondStereo::STEREOANY.

From CXSMILES

An initial stereochemistry assignment is done following the SMILES rules (see above).

A w: (wiggly bond) specification will set the stereochemistry of the start atom to ChiralType::CHI_UNSPECIFIED and double bonds to BondStereo::STEREOANY. Stereochemistry of ring bonds can be set using t, c, or ctu.

If 2D coordinates are present in the CXSMILES, atomic stereo can be set using `wU` or `wD` to create wedged bonds.

If 3D coordinates are present in the CXSMILES, they are used to set the stereochemistry for stereogenic atoms and bonds. This supersedes other specifications in the CXSMILES except for ctu and w.

Support for non-tetrahedral atomic stereochemistry

Starting with the 2022.09 release, the RDKit has partial, but evolving, support for non-tetrahedral stereochemistry. The status of this work is being tracked in this github issue: https://github.com/rdkit/rdkit/issues/4851

This code is being released in a preliminary state in order to get feedback as soon as we can and to start to gather experience working with these systems.

Status as of 2022.09.1 release

“Complete”

(Note that since is new territory, the term “complete” should be taken with a grain of salt.)

  • The basic representation

  • Parsing SMILES and SMARTS

  • Generation of 2D coordinates

  • Assignment of non-tetrahedral stereo from 3D structures

Partial
  • Writing SMILES. The SMILES generated should be correct, but they are not canonical.

  • Generation of 3D coordinates. The basics here work but the “chirality” of TBP and OH structures is not correct.

  • Writing mol files. Need wedged bonds for these to actually be done

Totally missing
  • Wedging bonds

  • Writing SMARTS

  • Substructure search integration

  • CIP assignment

  • Canonicalization

  • Stereochemistry cleanup: recognizing incorrect stereochemistry specifications

  • Assignment of non-tetrahedral stereo from 2D structures

SMILES notation

This discussion of the SMILES notation is drawn heavily from the OpenSMILES documentation: http://opensmiles.org/opensmiles.html Many thanks to the team which put that document together and to John Mayfield for his excellent CDK Depict tool, which I used double check my work on this.

The representation has a tag for what the stereo is, e.g. @SP, and a permutation number.

Square planar

@SP1

@SP2

@SP3

nts_sp1

nts_sp2

nts_sp3

U

4

Z

nts_sp4

Here are the ligand numberings for the 3 possible permutations of the sample molecule:

Label

A

B

C

D

SMILES

@SP1

0

1

2

3

C[Pt@SP1](F)(Cl)[H]

@SP2

0

2

1

3

C[Pt@SP2](Cl)(F)[H]

@SP3

0

1

3

2

C[Pt@SP3](F)([H])Cl

Trigonal bipyramidal

Here’s a specific example (from the OpenSMILES docs):

nts_tb2

Here are the ligand labels and the ligand numbering for @TB1:

nts_tb1

And then the ligand numberings for the 20 possible permutations of the sample molecule:

Label

A

B

C

D

E

SMILES

@TB1

0

4

1

2

3

S[As@TB1](F)(Cl)(Br)N

@TB2

0

4

1

3

2

S[As@TB2](F)(Br)(Cl)N

@TB3

0

3

1

2

4

S[As@TB3](F)(Cl)(N)Br

@TB4

0

3

1

4

2

S[As@TB4](F)(Br)(N)Cl

@TB5

0

2

1

3

4

S[As@TB5](F)(N)(Cl)Br

@TB6

0

2

1

4

3

S[As@TB6](F)(N)(Br)Cl

@TB7

0

1

2

3

4

S[As@TB7](N)(F)(Cl)Br

@TB8

0

1

2

4

3

S[As@TB8](N)(F)(Br)Cl

@TB9

1

4

0

2

3

F[As@TB9](S)(Cl)(Br)N

@TB11

1

4

0

3

2

F[As@TB11](S)(Br)(Cl)N

@TB10

1

3

0

2

4

F[As@TB10](S)(Cl)(N)Br

@TB12

1

3

0

4

2

F[As@TB12](S)(Br)(N)Cl

@TB13

1

2

0

3

4

F[As@TB13](S)(N)(Cl)Br

@TB14

1

2

0

4

3

F[As@TB14](S)(N)(Br)Cl

@TB15

2

4

0

1

3

F[As@TB15](Cl)(S)(Br)N

@TB20

2

4

0

3

1

F[As@TB20](Br)(S)(Cl)N

@TB16

2

3

0

1

4

F[As@TB16](Cl)(S)(N)Br

@TB19

2

3

0

4

1

F[As@TB19](Br)(S)(N)Cl

@TB17

3

4

0

1

2

F[As@TB17](Cl)(Br)(S)N

@TB18

3

4

0

2

1

F[As@TB18](Br)(Cl)(S)N

Octahedral

Here’s a specific example (an invented molecule):

nts_oh2

Here are the ligand labels and the ligand numbering for @OH1:

nts_oh1

And then the square planar shape and ligand numberings for the 30 possible permutations of the sample molecule:

Label

SP

A

B

C

D

E

F

SMILES

@OH1

U

0

5

1

2

3

4

O[Co@OH1](Cl)(C)(N)(F)P

@OH2

U

0

5

1

4

3

2

O[Co@OH2](Cl)(F)(N)(C)P

@OH3

U

0

4

1

2

3

5

O[Co@OH3](Cl)(C)(N)(P)F

@OH16

U

0

4

1

5

3

2

O[Co@OH16](Cl)(F)(N)(P)C

@OH6

U

0

3

1

2

4

5

O[Co@OH6](Cl)(C)(P)(N)F

@OH18

U

0

3

1

5

4

2

O[Co@OH18](Cl)(F)(P)(N)C

@OH19

U

0

2

1

3

4

5

O[Co@OH19](Cl)(P)(C)(N)F

@OH24

U

0

2

1

5

4

3

O[Co@OH24](Cl)(P)(F)(N)C

@OH25

U

0

1

2

3

4

5

O[Co@OH25](P)(Cl)(C)(N)F

@OH30

U

0

1

2

5

4

3

O[Co@OH30](P)(Cl)(F)(N)C

@OH4

Z

0

5

1

2

4

3

O[Co@OH4](Cl)(C)(F)(N)P

@OH14

Z

0

5

1

3

4

2

O[Co@OH14](Cl)(F)(C)(N)P

@OH5

Z

0

4

1

2

5

3

O[Co@OH5](Cl)(C)(F)(P)N

@OH15

Z

0

4

1

3

5

2

O[Co@OH15](Cl)(F)(C)(P)N

@OH7

Z

0

3

1

2

5

4

O[Co@OH7](Cl)(C)(P)(F)N

@OH17

Z

0

3

1

4

5

2

O[Co@OH17](Cl)(F)(P)(C)N

@OH20

Z

0

2

1

3

5

4

O[Co@OH20](Cl)(P)(C)(F)N

@OH23

Z

0

2

1

4

5

3

O[Co@OH23](Cl)(P)(F)(C)N

@OH26

Z

0

1

2

3

5

4

O[Co@OH26](P)(Cl)(C)(F)N

@OH29

Z

0

1

2

4

5

3

O[Co@OH29](P)(Cl)(F)(C)N

@OH10

4

0

5

1

4

2

3

O[Co@OH10](Cl)(N)(F)(C)P

@OH8

4

0

5

1

3

2

4

O[Co@OH8](Cl)(N)(C)(F)P

@OH11

4

0

4

1

5

2

3

O[Co@OH11](Cl)(N)(F)(P)C

@OH9

4

0

4

1

3

2

5

O[Co@OH9](Cl)(N)(C)(P)F

@OH13

4

0

3

1

4

2

4

O[Co@OH13](Cl)(N)(P)(F)C

@OH12

4

0

3

1

4

2

5

O[Co@OH12](Cl)(N)(P)(C)F

@OH22

4

0

2

1

5

3

4

O[Co@OH22](Cl)(P)(N)(F)C

@OH21

4

0

2

1

4

3

5

O[Co@OH21](Cl)(P)(N)(C)F

@OH28

4

0

1

2

5

3

4

O[Co@OH28](P)(Cl)(N)(F)C

@OH27

4

0

1

2

4

3

5

O[Co@OH27](P)(Cl)(N)(C)F

Duplicate ligands

One of the major differences between non-tetrahedral stereochemistry and the tetrahedral variant is that it’s possible to have non-tetrahedral stereo with central atoms which have duplicate ligands.

The classic example of this is cis-platin - Cl[Pt@SP1](Cl)(<-[NH3])<-[NH3] - vs trans-platin - Cl[Pt@SP2](Cl)(<-[NH3])<-[NH3] -

nts_ex1

nts_ex2

Cl[Pt@SP1](Cl)(<-[NH3])<-[NH3]

Cl[Pt@SP2](Cl)(<-[NH3])<-[NH3]

Treatment of implicit Hs

Implicit Hs are treated the same as in tetrahedral stereo: as if they are the first neighbors after the central atom. So the two smiles C[Pt@SP1H](Cl)F and C[Pt@SP1]([H])(Cl)F corresponds to the same structure.

This also works with multiple implicit Hs: C[Pt@SP1H2]Cl and C[Pt@SP1]([H])([H])Cl are equivalent.

Missing ligands

Coordination environments with missing ligands are treated as if the missing ligands were at the end of the ligand ordering. For example, this invented complex can be presented with the SMILES O[Mn@OH1](Cl)(C)(N)F.

nts_missing1

Compare this to the SMILES for the related complex shown above in the discussion of @OH stereo.

Chemical Reaction Handling

Reaction SMARTS

Not SMIRKS [1] , not reaction SMILES [2], derived from SMARTS [3].

The general grammar for a reaction SMARTS is :

reaction  ::=   reactants ">>" products
reactants ::=  molecules
products  ::=   molecules
molecules ::=  molecule
               molecules "." molecule
molecule  ::=   a valid SMARTS string without "." characters
                "(" a valid SMARTS string without "." characters ")"

Some features

Mapped dummy atoms in the product template are replaced by the corresponding atom in the reactant:

>>> from rdkit.Chem import AllChem
>>> rxn = AllChem.ReactionFromSmarts('[C:1]=[O,N:2]>>[C:1][*:2]')
>>> [Chem.MolToSmiles(x,1) for x in rxn.RunReactants((Chem.MolFromSmiles('CC=O'),))[0]]
['CCO']
>>> [Chem.MolToSmiles(x,1) for x in rxn.RunReactants((Chem.MolFromSmiles('CC=N'),))[0]]
['CCN']

but unmapped dummy atoms are left as dummies:

>>> rxn = AllChem.ReactionFromSmarts('[C:1]=[O,N:2]>>*[C:1][*:2]')
>>> [Chem.MolToSmiles(x,1) for x in rxn.RunReactants((Chem.MolFromSmiles('CC=O'),))[0]]
['*C(C)O']

“Any” bonds in the products are replaced by the corresponding bond in the reactant:

>>> rxn = AllChem.ReactionFromSmarts('[C:1]~[O,N:2]>>*[C:1]~[*:2]')
>>> [Chem.MolToSmiles(x,1) for x in rxn.RunReactants((Chem.MolFromSmiles('C=O'),))[0]]
['*C=O']
>>> [Chem.MolToSmiles(x,1) for x in rxn.RunReactants((Chem.MolFromSmiles('CO'),))[0]]
['*CO']
>>> [Chem.MolToSmiles(x,1) for x in rxn.RunReactants((Chem.MolFromSmiles('C#N'),))[0]]
['*C#N']

Intramolecular reactions can be expressed flexibly by including reactants in parentheses. This is demonstrated in this ring-closing metathesis example [5]:

>>> rxn = AllChem.ReactionFromSmarts("([C:1]=[C;H2].[C:2]=[C;H2])>>[*:1]=[*:2]")
>>> m1 = Chem.MolFromSmiles('C=CCOCC=C')
>>> ps = rxn.RunReactants((m1,))
>>> Chem.MolToSmiles(ps[0][0])
'C1=CCOC1'

Chirality

This section describes how chirality information in the reaction defition is handled. A consistent example, esterification of secondary alcohols, is used throughout [6].

If no chiral information is present in the reaction definition, the stereochemistry of the reactants is preserved, as is membership in enhanced stereo groups:

>>> alcohol1 = Chem.MolFromSmiles('CC(CCN)O')
>>> alcohol2 = Chem.MolFromSmiles('C[C@H](CCN)O')
>>> alcohol3 = Chem.MolFromSmiles('C[C@@H](CCN)O')
>>> acid = Chem.MolFromSmiles('CC(=O)O')
>>> rxn = AllChem.ReactionFromSmarts('[CH1:1][OH:2].[OH][C:3]=[O:4]>>[C:1][O:2][C:3]=[O:4]')
>>> ps=rxn.RunReactants((alcohol1,acid))
>>> Chem.MolToSmiles(ps[0][0],True)
'CC(=O)OC(C)CCN'
>>> ps=rxn.RunReactants((alcohol2,acid))
>>> Chem.MolToSmiles(ps[0][0],True)
'CC(=O)O[C@H](C)CCN'
>>> ps=rxn.RunReactants((alcohol3,acid))
>>> Chem.MolToSmiles(ps[0][0],True)
'CC(=O)O[C@@H](C)CCN'

You get the same result (retention of stereochemistry) if a mapped atom has the same chirality in both reactants and products:

>>> rxn = AllChem.ReactionFromSmarts('[C@H1:1][OH:2].[OH][C:3]=[O:4]>>[C@:1][O:2][C:3]=[O:4]')
>>> ps=rxn.RunReactants((alcohol1,acid))
>>> Chem.MolToSmiles(ps[0][0],True)
'CC(=O)OC(C)CCN'
>>> ps=rxn.RunReactants((alcohol2,acid))
>>> Chem.MolToSmiles(ps[0][0],True)
'CC(=O)O[C@H](C)CCN'
>>> ps=rxn.RunReactants((alcohol3,acid))
>>> Chem.MolToSmiles(ps[0][0],True)
'CC(=O)O[C@@H](C)CCN'

A mapped atom with different chirality in reactants and products leads to inversion of stereochemistry:

>>> rxn = AllChem.ReactionFromSmarts('[C@H1:1][OH:2].[OH][C:3]=[O:4]>>[C@@:1][O:2][C:3]=[O:4]')
>>> ps=rxn.RunReactants((alcohol1,acid))
>>> Chem.MolToSmiles(ps[0][0],True)
'CC(=O)OC(C)CCN'
>>> ps=rxn.RunReactants((alcohol2,acid))
>>> Chem.MolToSmiles(ps[0][0],True)
'CC(=O)O[C@@H](C)CCN'
>>> ps=rxn.RunReactants((alcohol3,acid))
>>> Chem.MolToSmiles(ps[0][0],True)
'CC(=O)O[C@H](C)CCN'

If a mapped atom has chirality specified in the reactants, but not in the products, the reaction destroys chirality at that center:

>>> rxn = AllChem.ReactionFromSmarts('[C@H1:1][OH:2].[OH][C:3]=[O:4]>>[C:1][O:2][C:3]=[O:4]')
>>> ps=rxn.RunReactants((alcohol1,acid))
>>> Chem.MolToSmiles(ps[0][0],True)
'CC(=O)OC(C)CCN'
>>> ps=rxn.RunReactants((alcohol2,acid))
>>> Chem.MolToSmiles(ps[0][0],True)
'CC(=O)OC(C)CCN'
>>> ps=rxn.RunReactants((alcohol3,acid))
>>> Chem.MolToSmiles(ps[0][0],True)
'CC(=O)OC(C)CCN'

And, finally, if chirality is specified in the products, but not the reactants, the reaction creates a stereocenter with the specified chirality:

>>> rxn = AllChem.ReactionFromSmarts('[CH1:1][OH:2].[OH][C:3]=[O:4]>>[C@:1][O:2][C:3]=[O:4]')
>>> ps=rxn.RunReactants((alcohol1,acid))
>>> Chem.MolToSmiles(ps[0][0],True)
'CC(=O)O[C@H](C)CCN'
>>> ps=rxn.RunReactants((alcohol2,acid))
>>> Chem.MolToSmiles(ps[0][0],True)
'CC(=O)O[C@H](C)CCN'
>>> ps=rxn.RunReactants((alcohol3,acid))
>>> Chem.MolToSmiles(ps[0][0],True)
'CC(=O)O[C@H](C)CCN'

This doesn’t make sense without including a bit more context around the stereocenter in the reaction definition:

>>> rxn = AllChem.ReactionFromSmarts('[CH3:5][CH1:1]([C:6])[OH:2].[OH][C:3]=[O:4]>>[C:5][C@:1]([C:6])[O:2][C:3]=[O:4]')
>>> ps=rxn.RunReactants((alcohol1,acid))
>>> Chem.MolToSmiles(ps[0][0],True)
'CC(=O)O[C@H](C)CCN'
>>> ps=rxn.RunReactants((alcohol2,acid))
>>> Chem.MolToSmiles(ps[0][0],True)
'CC(=O)O[C@H](C)CCN'
>>> ps=rxn.RunReactants((alcohol3,acid))
>>> Chem.MolToSmiles(ps[0][0],True)
'CC(=O)O[C@H](C)CCN'

Note that the chirality specification is not being used as part of the query: a molecule with no chirality specified can match a reactant with specified chirality.

In general, the reaction machinery tries to preserve as much stereochemistry information as possible. This works when a single new bond is formed to a chiral center:

>>> rxn = AllChem.ReactionFromSmarts('[C:1][C:2]-O>>[C:1][C:2]-S')
>>> alcohol2 = Chem.MolFromSmiles('C[C@@H](O)CCN')
>>> ps=rxn.RunReactants((alcohol2,))
>>> Chem.MolToSmiles(ps[0][0],True)
'C[C@@H](S)CCN'

But it fails if two or more bonds are formed:

>>> rxn = AllChem.ReactionFromSmarts('[C:1][C:2](-O)-F>>[C:1][C:2](-S)-Cl')
>>> alcohol = Chem.MolFromSmiles('C[C@@H](O)F')
>>> ps=rxn.RunReactants((alcohol,))
>>> Chem.MolToSmiles(ps[0][0],True)
'CC(S)Cl'

In this case, there’s just not sufficient information present to allow the information to be preserved. You can help by providing mapping information:

Some caveats We made this code as robust as we can, but this is a non-trivial problem and it’s certainly possible to get surprising results.

Things get tricky if atom ordering around a chiral center changes in the reaction SMARTS. Here are some of the situations that are currently handled correctly.

Reordering of the neighbors, but the number and atom mappings of neighbors remains constant. In this case there is no inversion of chirality even though the chiral tag on the chiral atom changes between the reactants and products:

>>> rxn = AllChem.ReactionFromSmarts('[C:1][C@:2]([F:3])[Br:4]>>[C:1][C@@:2]([S:4])[F:3]')
>>> mol = Chem.MolFromSmiles('C[C@@H](F)Br')
>>> ps=rxn.RunReactants((mol,))
>>> Chem.MolToSmiles(ps[0][0],True)
'C[C@@H](F)S'

Adding a neighbor to a chiral atom.

>>> rxn = AllChem.ReactionFromSmarts('[C:1][C@H:2]([F:3])[Br:4]>>[C:1][C@@:2](O)([F:3])[Br:4]')
>>> mol = Chem.MolFromSmiles('C[C@@H](F)Br')
>>> ps=rxn.RunReactants((mol,))
>>> Chem.MolToSmiles(ps[0][0],True)
'C[C@](O)(F)Br'

Removing a neighbor from a chiral atom.

>>> rxn = AllChem.ReactionFromSmarts('[C:1][C@:2](O)([F:3])[Br:4]>>[C:1][C@@H:2]([F:3])[Br:4]')
>>> mol = Chem.MolFromSmiles('C[C@@](O)(F)Br')
>>> ps=rxn.RunReactants((mol,))
>>> Chem.MolToSmiles(ps[0][0],True)
'C[C@H](F)Br'

Rules and warnings

  1. Include atom map information at the end of an atom query. So do [C,N,O:1] or [C;R:1].

  2. Don’t forget that unspecified bonds in SMARTS are either single or aromatic. Bond orders in product templates are assigned when the product template itself is constructed and it’s not always possible to tell if the bond should be single or aromatic:

>>> rxn = AllChem.ReactionFromSmarts('[#6:1][#7,#8:2]>>[#6:1][#6:2]')
>>> [Chem.MolToSmiles(x,1) for x in rxn.RunReactants((Chem.MolFromSmiles('C1NCCCC1'),))[0]]
['C1CCCCC1']
>>> [Chem.MolToSmiles(x,1) for x in rxn.RunReactants((Chem.MolFromSmiles('c1ncccc1'),))[0]]
['c1ccccc-1']

So if you want to copy the bond order from the reactant, use an “Any” bond:
>>> rxn = AllChem.ReactionFromSmarts('[#6:1][#7,#8:2]>>[#6:1]~[#6:2]')
>>> [Chem.MolToSmiles(x,1) for x in rxn.RunReactants((Chem.MolFromSmiles('c1ncccc1'),))[0]]
['c1ccccc1']

The Feature Definition File Format

An FDef file contains all the information needed to define a set of chemical features. It contains definitions of feature types that are defined from queries built up using Daylight’s SMARTS language. [3] The FDef file can optionally also include definitions of atom types that are used to make feature definitions more readable.

Chemical Features

Chemical features are defined by a Feature Type and a Feature Family. The Feature Family is a general classification of the feature (such as “Hydrogen-bond Donor” or “Aromatic”) while the Feature Type provides additional, higher-resolution, information about features. Pharmacophore matching is done using Feature Family’s. Each feature type contains the following pieces of information:

  • A SMARTS pattern that describes atoms (one or more) matching the feature type.

  • Weights used to determine the feature’s position based on the positions of its defining atoms.

Syntax of the FDef file

AtomType definitions

An AtomType definition allows you to assign a shorthand name to be used in place of a SMARTS string defining an atom query. This allows FDef files to be made much more readable. For example, defining a non-polar carbon atom like this:

AtomType Carbon_NonPolar [C&!$(C=[O,N,P,S])&!$(C#N)]

creates a new name that can be used anywhere else in the FDef file that it would be useful to use this SMARTS. To reference an AtomType, just include its name in curly brackets. For example, this excerpt from an FDef file defines another atom type - Hphobe - which references the Carbon_NonPolar definition:

AtomType Carbon_NonPolar [C&!$(C=[O,N,P,S])&!$(C#N)]
AtomType Hphobe [{Carbon_NonPolar},c,s,S&H0&v2,F,Cl,Br,I]

Note that {Carbon_NonPolar} is used in the new AtomType definition without any additional decoration (no square brackes or recursive SMARTS markers are required).

Repeating an AtomType results in the two definitions being combined using the SMARTS “,” (or) operator. Here’s an example:

AtomType d1 [N&!H0]
AtomType d1 [O&!H0]

This is equivalent to:

AtomType d1 [N&!H0,O&!H0]

Which is equivalent to the more efficient:

AtomType d1 [N,O;!H0]

Note that these examples tend to use SMARTS’s high-precedence and operator “&” and not the low-precedence and “;”. This can be important when AtomTypes are combined or when they are repeated. The SMARTS “,” operator is higher precedence than “;”, so definitions that use “;” can lead to unexpected results.

It is also possible to define negative AtomType queries:

AtomType d1 [N,O,S]
AtomType !d1 [H0]

The negative query gets combined with the first to produce a definition identical to this:

AtomType d1 [!H0;N,O,S]

Note that the negative AtomType is added to the beginning of the query.

Feature definitions

A feature definition is more complex than an AtomType definition and stretches across multiple lines:

DefineFeature HDonor1 [N,O;!H0]
Family HBondDonor
Weights 1.0
EndFeature

The first line of the feature definition includes the feature type and the SMARTS string defining the feature. The next two lines (order not important) define the feature’s family and its atom weights (a comma-delimited list that is the same length as the number of atoms defining the feature). The atom weights are used to calculate the feature’s locations based on a weighted average of the positions of the atom defining the feature. More detail on this is provided below. The final line of a feature definition must be EndFeature. It is perfectly legal to mix AtomType definitions with feature definitions in the FDef file. The one rule is that AtomTypes must be defined before they are referenced.

Additional syntax notes:

  • Any line that begins with a # symbol is considered a comment and will be ignored.

  • A backslash character, , at the end of a line is a continuation character, it indicates that the data from that line is continued on the next line of the file. Blank space at the beginning of these additional lines is ignored. For example, this AtomType definition:

    AtomType tButylAtom [$([C;!R](-[CH3])(-[CH3])(-[CH3])),\
    $([CH3](-[C;!R](-[CH3])(-[CH3])))]
    

    is exactly equivalent to this one:

    AtomType tButylAtom [$([C;!R](-[CH3])(-[CH3])(-[CH3])),$([CH3](-[C;!R](-[CH3])(-[CH3])))]
    

    (though the first form is much easier to read!)

Atom weights and feature locations

Frequently Asked Question(s)

  • What happens if a Feature Type is repeated in the file? Here’s an example:

    DefineFeature HDonor1 [O&!H0]
    Family HBondDonor
    Weights 1.0
    EndFeature
    
    DefineFeature HDonor1 [N&!H0]
    Family HBondDonor
    Weights 1.0
    EndFeature
    

    In this case both definitions of the HDonor1 feature type will be active. This is functionally identical to:

    DefineFeature HDonor1 [O,N;!H0]
    Family HBondDonor
    Weights 1.0
    EndFeature
    

    However the formulation of this feature definition with a duplicated feature type is considerably less efficient and more confusing than the simpler combined definition.

Representation of Pharmacophore Fingerprints

In the RDKit scheme the bit ids in pharmacophore fingerprints are not hashed: each bit corresponds to a particular combination of features and distances. A given bit id can be converted back to the corresponding feature types and distances to allow interpretation. An illustration for 2D pharmacophores is shown in Figure 1: Bit numbering in pharmacophore fingerprints.

_images/picture_10.jpg

Figure 1: Bit numbering in pharmacophore fingerprints

Atom-Atom Matching in Substructure Queries

When doing substructure matches for queries derived from SMARTS the rules for which atoms in the molecule should match which atoms in the query are well defined.[#smarts]_ The same is not necessarily the case when the query molecule is derived from a mol block or SMILES.

The general rule used in the RDKit is that if you don’t specify a property in the query, then it’s not used as part of the matching criteria and that Hs are ignored. This leads to the following behavior:

Molecule

Query

Match

CCO

CCO

Yes

CC[O-]

CCO

Yes

CCO

CC[O-]

No

CC[O-]

CC[O-]

Yes

CC[O-]

CC[OH]

Yes

CCOC

CC[OH]

Yes

CCOC

CCO

Yes

CCC

CCC

Yes

CC[14C]

CCC

Yes

CCC

CC[14C]

No

CC[14C]

CC[14C]

Yes

OCO

C

Yes

OCO

[CH]

No

OCO

[CH2]

No

OCO

[CH3]

No

OCO

O[CH3]

Yes

O[CH2]O

C

Yes

O[CH2]O

[CH2]

No

Demonstrated here:

>>> Chem.MolFromSmiles('CCO').HasSubstructMatch(Chem.MolFromSmiles('CCO'))
True
>>> Chem.MolFromSmiles('CC[O-]').HasSubstructMatch(Chem.MolFromSmiles('CCO'))
True
>>> Chem.MolFromSmiles('CCO').HasSubstructMatch(Chem.MolFromSmiles('CC[O-]'))
False
>>> Chem.MolFromSmiles('CC[O-]').HasSubstructMatch(Chem.MolFromSmiles('CC[O-]'))
True
>>> Chem.MolFromSmiles('CC[O-]').HasSubstructMatch(Chem.MolFromSmiles('CC[OH]'))
True
>>> Chem.MolFromSmiles('CCOC').HasSubstructMatch(Chem.MolFromSmiles('CC[OH]'))
True
>>> Chem.MolFromSmiles('CCOC').HasSubstructMatch(Chem.MolFromSmiles('CCO'))
True
>>> Chem.MolFromSmiles('CCC').HasSubstructMatch(Chem.MolFromSmiles('CCC'))
True
>>> Chem.MolFromSmiles('CC[14C]').HasSubstructMatch(Chem.MolFromSmiles('CCC'))
True
>>> Chem.MolFromSmiles('CCC').HasSubstructMatch(Chem.MolFromSmiles('CC[14C]'))
False
>>> Chem.MolFromSmiles('CC[14C]').HasSubstructMatch(Chem.MolFromSmiles('CC[14C]'))
True
>>> Chem.MolFromSmiles('OCO').HasSubstructMatch(Chem.MolFromSmiles('C'))
True
>>> Chem.MolFromSmiles('OCO').HasSubstructMatch(Chem.MolFromSmiles('[CH]'))
False
>>> Chem.MolFromSmiles('OCO').HasSubstructMatch(Chem.MolFromSmiles('[CH2]'))
False
>>> Chem.MolFromSmiles('OCO').HasSubstructMatch(Chem.MolFromSmiles('[CH3]'))
False
>>> Chem.MolFromSmiles('OCO').HasSubstructMatch(Chem.MolFromSmiles('O[CH3]'))
True
>>> Chem.MolFromSmiles('O[CH2]O').HasSubstructMatch(Chem.MolFromSmiles('C'))
True
>>> Chem.MolFromSmiles('O[CH2]O').HasSubstructMatch(Chem.MolFromSmiles('[CH2]'))
False

Generic (“Markush”) queries in substructure matching

Note This section describes functionality added in the 2022.03.1 release of the RDKit.

The RDKit supports a set of generic queries used as part of the Beilstein and Reaxys systems. Here’s an example:

_images/ary_group.png

Information about generic queries can be read in from CXSMILES or V3000 Mol blocks (as SUP SGroups) and then calling the function Chem.SetGenericQueriesFromProperties() with the molecule to be modified as an argument. These features are not used by default when doing substructure queries, but can be enabled by setting the option SubstructMatchParameters.useGenericMatchers to True

Here’s an example of using the features:

>>> q = Chem.MolFromSmarts('OC* |$;;ARY$|')
>>> Chem.SetGenericQueriesFromProperties(q)
>>> Chem.MolFromSmiles('C1CCCCC1CO').HasSubstructMatch(q)
True
>>> Chem.MolFromSmiles('c1ccccc1CO').HasSubstructMatch(q)
True
>>> ps = Chem.SubstructMatchParameters()
>>> ps.useGenericMatchers = True
>>> Chem.MolFromSmiles('C1CCCCC1CO').HasSubstructMatch(q,ps)
False
>>> Chem.MolFromSmiles('c1ccccc1CO').HasSubstructMatch(q,ps)
True

Here are the supported groups and a brief description of what they mean:

Alkyl (ALK)

alkyl side chains (not an H atom)

AlkylH (ALH)

alkyl side chains including an H atom

Alkenyl (AEL)

alkenyl side chains

AlkenylH (AEH)

alkenyl side chains or an H atom

Alkynyl (AYL)

alkynyl side chains

AlkynylH (AYH)

alkynyl side chains or an H atom

Alkoxy (AOX)

alkoxy side chains

AlkoxyH (AOH)

alkoxy side chains or an H atom

Carbocyclic (CBC)

carbocyclic side chains

CarbocyclicH (CBH)

carbocyclic side chains or an H atom

Carbocycloalkyl (CAL)

cycloalkyl side chains

CarbocycloalkylH (CAH)

cycloalkyl side chains or an H atom

Carbocycloalkenyl (CEL)

cycloalkenyl side chains

CarbocycloalkenylH (CEH)

cycloalkenyl side chains or an H atom

Carboaryl (ARY)

all-carbon aryl side chains

CarboarylH (ARH)

all-carbon aryl side chains or an H atom

Cyclic (CYC)

cyclic side chains

CyclicH (CYH)

cyclic side chains or an H atom

Acyclic(ACY)

acyclic side chains (not an H atom)

AcyclicH (ACH)

acyclic side chains or an H atom

Carboacyclic (ABC)

all-carbon acyclic side chains

CarboacyclicH (ABH)

all-carbon acyclic side chains or an H atom

Heteroacyclic (AHC)

acyclic side chains with at least one heteroatom

HeteroacyclicH (AHH)

acyclic side chains with at least one heteroatom or an H atom

Heterocyclic (CHC)

cyclic side chains with at least one heteroatom

HeterocyclicH (CHH)

cyclic side chains with at least one heteroatom or an H atom

Heteroaryl (HAR)

aryl side chains with at least one heteroatom

HeteroarylH (HAH)

aryl side chains with at least one heteroatom or an H atom

NoCarbonRing (CXX)

ring containing no carbon atoms

NoCarbonRingH (CXH)

ring containing no carbon atoms or an H atom

Group (G)

any group (not H atom)

GroupH (GH)

any group (including H atom)

Group* (G*)

any group with a ring closure

GroupH* (GH*)

any group with a ring closure or an H atom

For more detailed descriptions, look at the documentation for the C++ file GenericGroups.h

Molecular Sanitization

The molecule parsing functions all, by default, perform a “sanitization” operation on the molecules read. The idea is to generate useful computed properties (like hybridization, ring membership, etc.) for the rest of the code and to ensure that the molecules are “reasonable”: that they can be represented with octet-complete Lewis dot structures.

Here are the steps involved, in order.

  1. clearComputedProps: removes any computed properties that already exist

    on the molecule and its atoms and bonds. This step is always performed.

  2. cleanUp: standardizes a small number of non-standard valence states. The clean up operations are:

    • Neutral 5 valent Ns with double bonds to Os are converted to the zwitterionic form. Example: N(=O)=O -> [N+](=O)O-]

    • Neutral 5 valent Ns with triple bonds to another N are converted to the zwitterionic form. Example: C-N=N#N -> C-N=[N+]=[N-]

    • Neutral 5 valent phosphorus with one double bond to an O and another to either a C or a P are converted to the zwitterionic form. Example: C=P(=O)O -> C=[P+]([O-])O

    • Neutral Cl, Br, or I with exclusively O neighbors, and a valence of 3, 5, or 7, are converted to the zwitterionic form. This covers things like chlorous acid, chloric acid, and perchloric acid. Example: O=Cl(=O)O -> [O-][Cl+2][O-]O

    This step should not generate exceptions.

  3. cleanUpOrganometallics: standardizes a small number of non-standard situations encountered in organometallics. The cleanup operations are:

    • replaces single bonds from hypervalent atoms to metals with dative bonds.

    This step should not generate exceptions.

  4. updatePropertyCache: calculates the explicit and implicit valences on all atoms. This generates exceptions for atoms in higher-than-allowed valence states. This step is always performed, but if it is “skipped” the test for non-standard valences will not be carried out.

  5. symmetrizeSSSR: calls the symmetrized smallest set of smallest rings algorithm (discussed in the Getting Started document).

  6. Kekulize: converts aromatic rings to their Kekule form. Will raise an exception if a ring cannot be kekulized or if aromatic bonds are found outside of rings.

  7. assignRadicals: determines the number of radical electrons (if any) on each atom.

  8. setAromaticity: identifies the aromatic rings and ring systems (see above), sets the aromatic flag on atoms and bonds, sets bond orders to aromatic.

  9. setConjugation: identifies which bonds are conjugated

  10. setHybridization: calculates the hybridization state of each atom

  11. cleanupChirality: removes chiral tags from atoms that are not sp3 hybridized.

  12. adjustHs: adds explicit Hs where necessary to preserve the chemistry. This is typically needed for heteroatoms in aromatic rings. The classic example is the nitrogen atom in pyrrole.

  13. updatePropertyCache: re-calculates the explicit and implicit valences on

all atoms. This generates exceptions for atoms in higher-than-allowed valence states. This step is required to catch some edge cases where input atoms with non-physical valences are accepted if they are flagged as aromatic.

The individual steps can be toggled on or off when calling MolOps::sanitizeMol or Chem.SanitizeMol.

Valence calculation and allowed valences

The RDKit is, by default, fairly strict in the way it enforces allowed valences when sanitizing structures (this is done during the updatePropertyCache step of sanitization): atoms which have an explicit valence (sum of the specified bond orders + specified H count) exceeding the maximum allowed valence for the element will raise an exception.

Allowed valences of the elements (as of 2024.09.1):

H 1

He 0

Li 1 -1

Be 2

B 3

C 4

N 3

O 2

F 1

Ne 0

Na 1,-1

Mg 2,-1

Al 3

Si 4

P 3,5

S 2,4,6

Cl 1

Ar 0

K 1,-1

Ca 2,-1

Ga 3

Ge 4

As 3,5

Se 2,4,6

Br 1

Kr 0

Rb 1,-1

Sr 2,-1

In 3

Sn 2,4

Sb 3,5

Te 2,4,6

I 1,3,5

Xe 0,2,4,6

Cs 1,-1

Ba 2,-1

Tl -1

Pb 2,4

Bi 3,5

Po 2,4,6

At 1,3,5

Rn 0

Elements not listed in the table have a valence of -1.

An allowed valence of -1 indicates that the element can have any valence value. Implicit Hs will not be added to atoms with a possible valence of -1 when the explicit valence exceeds the highest specified value. So, for example, an Mg atom with a single bond to it (explicit valence = 1) will have one implicit H added to it, while an Mg atom with three bonds to it will have no implicit Hs added. Atoms where the only allowed valence is -1 will never have implicit Hs added.

The allowed valences of charged atoms are calculated by looking at the isoelectronic element’s allowed valences. For example, N+ has the same allowed valences as C, while N- has the same allowed valences as O. P-2, S-, As-2, and Se- are special cases: they all have allowed valences of 1, 3 and 5.

JSON Support

The RDKit supports writing to/reading from two closely related JSON formats: commonchem (https://github.com/CommonChem/CommonChem) and rdkitjson. commonchem is a well-documented format designed to be used for efficient interchange between molecular toolkits. rdkitjson is an extension to commonchem which includes additional features allowing RDKit molecules to be serialized to JSON. The extensions in rdkitjson - enhanced stereo and substance groups - are generally useful, so it’s easy to imagine them being integrated into commonchem at some point in the future.

Lists of molecules can be converted to JSON with MolInterchange::MolsToJSONData() (C++) or Chem.MolsToJSONData() (Python). Those calls take an optional parameters object which can be used to specify whether commonchem or rdkitjson is generated. The default is to generate rdkitjson.

JSON data can be converted back to RDKit molecules using MolInterchange::JSONDataToMols() (C++) or Chem.JSONDataToMols() (Python). The parser will automatically determine whether or not its working with commonchem or rdkitjson.

rdkitjson format

Enhanced stereo

Here’s the rdkitjson representation of the stereo groups from the molecule C[C@@H]1C([C@H](O)F)O[C@H](C)C([C@@H](O)F)[C@@H]1C |a:7,o1:3,10,&1:1,&2:13|:

'stereoGroups': [{'type': 'abs', 'atoms': [7]},
 {'type': 'or', 'atoms': [3, 10]},
 {'type': 'and', 'atoms': [1]},
 {'type': 'and', 'atoms': [13]}],

Substance groups

Here’s the rdkitjson representation of a SUP substance group:

'substanceGroups': [{'properties': {'TYPE': 'SUP',
   'index': 1,
   'LABEL': 'Boc',
   'DATAFIELDS': '[]'},
  'atoms': [7, 8, 9, 10, 11, 12, 13],
  'bonds': [8],
  'brackets': [[[6.24, -2.9, 0.0], [6.24, -2.9, 0.0], [0.0, 0.0, 0.0]]],
  'cstates': [{'bond': 8, 'vector': [0.0, 0.82, 0.0]}],
  'attachPoints': [{'aIdx': 12, 'lvIdx': 5, 'id': '1'}]}],

and one for an SRU group:

'substanceGroups': [{'properties': {'TYPE': 'SRU',
   'index': 1,
   'CONNECT': 'HT',
   'LABEL': 'n',
   'DATAFIELDS': '[]'},
  'atoms': [2, 1, 4],
  'bonds': [2, 0],
  'brackets': [[[-3.9538, 4.3256, 0.0],
                [-3.0298, 2.7252, 0.0],
                [0.0, 0.0, 0.0]],
               [[-5.4618, 2.8611, 0.0],
                [-6.3858, 4.4615, 0.0],
                [0.0, 0.0, 0.0]]]}],

Implementation Details

“Magic” Property Values

The following property values are regularly used in the RDKit codebase and may be useful to client code.

ROMol (Mol in Python)

Property Name

Use

MolFileComments

Read from/written to the comment line of CTABs.

MolFileInfo

Read from/written to the info line of CTABs.

_MolFileChiralFlag

Read from/written to the chiral flag of CTABs.

_Name

Read from/written to the name line of CTABs.

_smilesAtomOutputOrder

The order in which atoms were written to SMILES

_smilesBondOutputOrder

The order in which bonds were written to SMILES

Atom

Property Name

Use

_CIPCode

the CIP code (R or S) of the atom

_CIPRank

the integer CIP rank of the atom

_ChiralityPossible

set if an atom is a possible chiral center

_MolFileRLabel

integer R group label for an atom, read from/written to CTABs.

_ReactionDegreeChanged

set on an atom in a product template of a reaction if its degree changes in the reaction

_protected

atoms with this property set will not be considered as matching reactant queries in reactions

dummyLabel

(on dummy atoms) read from/written to CTABs as the atom symbol

molAtomMapNumber

the atom map number for an atom, read from/written to SMILES and CTABs

molfileAlias

the mol file alias for an atom (follows A tags), read from/written to CTABs

molFileValue

the mol file value for an atom (follows V tags), read from/written to CTABs

molFileInversionFlag

used to flag whether stereochemistry at an atom changes in a reaction, read from/written to CTABs, determined automatically from SMILES

molRxnComponent

which component of a reaction an atom belongs to, read from/written to CTABs

molRxnRole

which role an atom plays in a reaction (1=Reactant, 2=Product, 3=Agent), read from/written to CTABs

smilesSymbol

determines the symbol that will be written to a SMILES for the atom

Thread safety and the RDKit

While writing the RDKit, we did attempt to ensure that the code would work in a multi-threaded environment by avoiding use of global variables, etc. However, making code thread safe is not a completely trivial thing, so there are no doubt some gaps. This section describes which pieces of the code base have explicitly been tested for thread safety.

Note: With the exception of the small number of methods/functions

that take a numThreads argument, this section does not apply to using the RDKit from Python threads. Boost.Python ensures that only one thread is calling into the C++ code at any point. To get concurrent execution in Python, use the multiprocessing module or one of the other standard python approaches for this .

What has been tested

  • Reading molecules from SMILES/SMARTS/Mol blocks

  • Writing molecules to SMILES/SMARTS/Mol blocks (see below)

  • Generating 2D coordinates

  • Generating 3D conformations with the distance geometry code

  • Optimizing molecules with UFF or MMFF

  • Generating fingerprints

  • The descriptor calculators in $RDBASE/Code/GraphMol/Descriptors

  • Substructure searching (Note: if a query molecule contains recursive queries, it may not be safe to use it concurrently on multiple threads, see below)

  • The Subgraph code

  • The ChemTransforms code

  • The chemical reactions code

  • The Open3DAlign code

  • The MolDraw2D drawing code

  • The InChI code, with InChI IUPAC v1.06

Known Problems

  • The MolSuppliers (e.g. SDMolSupplier, SmilesMolSupplier?) change their internal state when a molecule is read. It is not safe to use one supplier on more than one thread.

  • Substructure searching using query molecules that include recursive queries. The recursive queries modify their internal state when a search is run, so it’s not safe to use the same query concurrently on multiple threads. If the code is built using the RDK_BUILD_THREADSAFE_SSS argument (the default for the binaries we provide), a mutex is used to ensure that only one thread is using a given recursive query at a time.

  • Calling MolToSmiles() on the same molecule from multiple threads can lead to data races with the calculated properties on the molecule.

Implementation of the TPSA Descriptor

The topological polar surface area (TPSA) descriptor implemented in the RDKit is described in a publication by Peter Ertl et al. (https://pubs.acs.org/doi/abs/10.1021/jm000942e) The RDKit’s implementation differs from what is described in that publication. This section describes the difference and why it’s there.

The RDKit’s TPSA implementation only includes, by default, contributions from N and O atoms. Table 1 of the TPSA publication. however, includes parameters for polar S and P in addition to N and O. What’s going on?

The original TPSA implementation that is in the Daylight Contrib dir (http://www.daylight.com/download/contrib/tpsa.html) does not include contributions from polar S or P and, it turns out, the reference values that are included in the TPSA paper also don’t include S or P contributions. For example, the TPSA provided in Table 3 for foscarnet (SMILES OC(=O)P(=O)(O)O), 94.8, corresponds the sum of the O contributions - 3x20.23 + 2*17.07 = 94.8. Adding the P contribution - 9.81- would give a PSA of 104.6. This is also true for the other P and S containing compounds in Table 3.

In the RDKit implementation, we chose to reproduce the behavior of the tpsa.c Contrib program and what is provided in Table 3 of the paper, so polar S and P are ignored. Based on a couple of user requests, for the 2018.09 release of the RDKit we added the option to include S and P contributions:

>>> from rdkit.Chem import Descriptors
>>> Descriptors.TPSA(Chem.MolFromSmiles('OC(=O)P(=O)(O)O')) # foscarnet
94.83
>>> Descriptors.TPSA(Chem.MolFromSmiles('OC(=O)P(=O)(O)O'), includeSandP=True)
104.64...
>>> Descriptors.TPSA(Chem.MolFromSmiles('Cc1ccccc1N1C(=O)c2cc(S(N)(=O)=O)c(Cl)cc2NC1C')) # metolazone
92.5
>>> Descriptors.TPSA(Chem.MolFromSmiles('Cc1ccccc1N1C(=O)c2cc(S(N)(=O)=O)c(Cl)cc2NC1C'), includeSandP=True)
100.88

Atom Properties and SDF files

Note This section describes functionality added in the 2019.03.1 release of the RDKit.

By default the rdkit.Chem.rdmolfiles.SDMolSupplier and rdkit.Chem.rdmolfiles.ForwardSDMolSupplier classes (RDKit::SDMolSupplier and RDKit::ForwardMolSupplier in C++) can now recognize some molecular properties as property lists and them into atomic properties. Properties with names that start with atom.prop, atom.iprop, atom.dprop, or atom.bprop are converted to atomic properties of type string, int (64 bit), double, or bool respectively.

Here’s a sample block from an SDF that demonstrates all of the features, they are explained below:

property_example
    RDKit  2D

  3  3  0  0  0  0  0  0  0  0999 V2000
    0.8660    0.0000    0.0000 C   0  0  0  0  0  0  0  0  0  0  0  0
  -0.4330    0.7500    0.0000 N   0  0  0  0  0  0  0  0  0  0  0  0
  -0.4330   -0.7500    0.0000 C   0  0  0  0  0  0  0  0  0  0  0  0
  1  2  1  0
  2  3  1  0
  3  1  1  0
M  END
>  <atom.dprop.PartialCharge>  (1)
0.008 -0.314 0.008

>  <atom.iprop.NumHeavyNeighbors>  (1)
2 2 2

>  <atom.prop.AtomLabel>  (1)
C1 N2 C3

>  <atom.bprop.IsCarbon>  (1)
1 0 1

>  <atom.prop.PartiallyMissing>  (1)
one n/a three

>  <atom.iprop.PartiallyMissingInt>  (1)
[?] 2 2 ?

$$$$

Every atom property list should contain a number of space-delimited elements equal to the number of atoms. Missing values are, by default, indicated with the string n/a. The missing value marker can be changed by beginning the property list with a value in square brackets. So, for example, the property PartiallyMissing is set to “one” for atom 0, “three” for atom 2, and is not set for atom 1. Similarly the property PartiallyMissingInt is set to 2 for atom 0, 2 for atom 1, and is not set for atom 2.

This behavior is enabled by default and can be turned on/off with the rdkit.Chem.rdmolfiles.SetProcessPropertyLists method.

If you have atom properties that you would like to have written to SDF files, you can use the functions rdkit.Chem.rdmolfiles.CreateAtomStringPropertyList(), rdkit.Chem.rdmolfiles.CreateAtomIntPropertyList(), rdkit.Chem.rdmolfiles.CreateAtomDoublePropertyList(), or rdkit.Chem.rdmolfiles.CreateAtomBoolPropertyList() :

>>> m = Chem.MolFromSmiles('CO')
>>> m.GetAtomWithIdx(0).SetDoubleProp('foo',3.14)
>>> Chem.CreateAtomDoublePropertyList(m,'foo')
>>> m.GetProp('atom.dprop.foo')
'3.1400000000000001 n/a'
>>> from io import StringIO
>>> sio = StringIO()
>>> w = Chem.SDWriter(sio)
>>> w.write(m)
>>> w=None
>>> print(sio.getvalue())   

     RDKit          2D

  2  1  0  0  0  0  0  0  0  0999 V2000
    0.0000    0.0000    0.0000 C   0  0  0  0  0  0  0  0  0  0  0  0
    1.2990    0.7500    0.0000 O   0  0  0  0  0  0  0  0  0  0  0  0
  1  2  1  0
M  END
>  <atom.dprop.foo>  (1)
3.1400000000000001 n/a

$$$$

Support for Enhanced Stereochemistry

Overview

Enhanced stereochemistry is used to indicate that a molecule represents more than one possible diastereomer. AND indicates that a molecule is a mixture of molecules. OR indicates unknown single substances, and ABS indicates a single substance. This follows, the convention used in V3k mol files: groups of atoms with specified stereochemistry with an ABS, AND, or OR marker indicating what is known.

Here are some illustrations of what the various combinations mean:

What’s drawn

Mixture?

What it means

and1_and2_base

mixture

and1_and2_expand

and1_cis_base

mixture

and1_cis_expand

and1_trans_base

mixture

and1_trans_expand

or1_or2_base

single

or1_or2_expand

or1_cis_base

single

or1_cis_expand

or1_trans_base

single

or1_trans_expand

abs_and_base

mixture

abs_and_expand

abs_or_base

single

abs_or_expand

Representation

Stored as a vector of rdkit.Chem.rdchem.StereoGroup objects on a molecule. Each StereoGroup keeps track of its type and the set of atoms that make it up.

Use cases

The initial target is to not lose data on an V3k mol -> RDKit -> V3k mol round trip. Manipulation and depiction are future goals.

It is possible to enumerate the elements of a StereoGroup using the function rdkit.Chem.EnumerateStereoisomers.EumerateStereoisomers(). Note that this removes the StereoGroup information from the products since they now correspond to specific molecules:

>>> m = Chem.MolFromSmiles('C[C@H](F)C[C@H](O)Cl |&1:1|')
>>> m.GetStereoGroups()[0].GetGroupType()
rdkit.Chem.rdchem.StereoGroupType.STEREO_AND
>>> [x.GetIdx() for x in m.GetStereoGroups()[0].GetAtoms()]
[1]
>>> from rdkit.Chem.EnumerateStereoisomers import EnumerateStereoisomers
>>> [Chem.MolToCXSmiles(x) for x in EnumerateStereoisomers(m)]
['C[C@@H](F)C[C@H](O)Cl', 'C[C@H](F)C[C@H](O)Cl']

Reactions also preserve StereoGroup``s. Product atoms are included in the ``StereoGroup as long as the reaction doesn’t create or destroy chirality at the atom.

>>> def clearAllAtomProps(mol):
...  """So that atom mapping isn't shown"""
...  for atom in mol.GetAtoms():
...   for key in atom.GetPropsAsDict():
...    atom.ClearProp(key)
...
>>> rxn = AllChem.ReactionFromSmarts('[C:1]F >> [C:1]Br')
>>> ps=rxn.RunReactants([m])
>>> clearAllAtomProps(ps[0][0])
>>> Chem.MolToCXSmiles(ps[0][0])
'C[C@H](Br)C[C@H](O)Cl |&1:1|'

Atropisomeric Bonds

Some single bonds have restricted rotation because of steric interactions between the groups on adjacent atoms. If the groups on the adjacent atoms are different from each other, chirality can be induced. An atropisomer bond is such a restricted rotation bond.

The requirements for a bond to be eligible for atropisomerism in the RDKit are:

  • The bond must be a single bond between SP2 hybridized atoms.

  • The neighboring bonds must be single, double or aromatic.

  • If there are two groups on either end, those groups must be different as per CIP rules.

  • Currently RDKit considers ring bonds as potential atropisomer bonds only if the ring in which the bond appears is 8 atoms or larger (thus allowing macrocycles).

  • The molecule must have coordinates for atropisomer bonds to be interpreted.

The definition of potential atropisomer bonds is based on the wedging of adjacent bonds.

Defining Atropisomers

At least one of the neighbor bonds of one of the atoms of the potential atropisomer bond must be a single or aromatic bond, and must have a bond direction that is either “wedged” or “hashed”. If any of the neighbor bonds is marked as “sqwiggly”, the bond is considered to have “Any” stereochemistry. Example structure:

_images/atrop_example1.png

If more than one of the neighbor bonds are wedged or hashed, they must be consistent.

For example, if two neighbor bonds on the same end atom are wedged/hashed, one must be a wedge and the other must be a hash. If neighbor bonds on different ends of the atropisomer bond are wedged, and the two bonds are opposite sides of the potential atropisomer bond (the dihedral angle is greater than 90 degrees or less than -90 degrees), the two must both be wedges or both must be hashed. If the two wedged/hashed neighbor bonds are on the same side of the potential atropisomer bond (the dihedral angle is less than 90 degrees and greater than -90 degrees), one must be a wedge and the other a hash.

Examples – valid atropisomers with multiple wedges:

_images/atrop_example2.png

Examples – invalid atropisomers with multiple wedges:

_images/atrop_example3.png

Note: the RDKit software makes no attempt to determine if the bond is actually rotationally constrained. If the bond meets the requirements above, it is marked as an atropisomer.

Internal Representation of Atropisomers

To help with the rest of the explanation, we include the bond indices in the molecule drawing:

_images/atrop_representation1.png

Here’s part of the Debug output for that molecule:

Atoms:
  ...
  1 6 C chg: 0  deg: 3 exp: 4 imp: 0 hyb: SP2 arom?: 1
  ...
  5 6 C chg: 0  deg: 3 exp: 4 imp: 0 hyb: SP2 arom?: 1
  ...
  7 6 C chg: 0  deg: 3 exp: 4 imp: 0 hyb: SP2 arom?: 1
  8 7 N chg: 0  deg: 3 exp: 3 imp: 0 hyb: SP2 arom?: 1
  9 6 C chg: 0  deg: 3 exp: 4 imp: 0 hyb: SP2 arom?: 1
  ...
  13 6 C chg: 0  deg: 3 exp: 4 imp: 0 hyb: SP2 arom?: 1
  ...
Bonds:
  ...
  6 5->7 order: a conj?: 1 aromatic?: 1
  7 7->8 order: 1 stereo: CCW bonds: (14 6 8 15) conj?: 1
  8 8->9 order: a conj?: 1 aromatic?: 1
  ...
  14 7->1 order: a dir: wedge conj?: 1 aromatic?: 1
  15 13->8 order: a conj?: 1 aromatic?: 1

This tells us that bond 7 is atropisomeric and that when looking down the bond (from atom C7 to atom N8), the rotation direction between bonds 14 and 8 (these are the bonds to the lowest numbered atoms on each of the atropisomeric bond) is counterclockwise (CCW).

Formats supporting atropisomers

Atropisomers are supported for molecule parsing and writing in Mol, MRV, and CXSmiles formats. For reactions, atropisomers are supported for in rxn, MRV and CXSmiles formats. Atropisomers can be parsed from a CDXML file.

Enhanced Stereochemistry

Atropisomers can be part of Enhanced Stereochemistry Groups (Or, And, or Absolute). This is indicated by marking one or both of the atropisomer bond’s atoms as being in the enhanced Stereo Group

Example:

_images/atrop_example4.png

3D Coordinates for Input of Atropisomers

If 3D coordinates are available (and 2D coordinates are not), the atropisomer bond is marked only if one of the neighbor bonds is wedged/hashed. The wedge/hash information is ignored except for signaling the presence of the atropisomer. The actual configuration is determined by the 3D coordinates.

Here’s an example. The drawing at the left shows a 3D structure with a wedged bond on one end of the atropisomer bond. The drawing at the right shows a 2D drawing of the same structure. In both cases, the atropisomer bond is highlighted in red. Notice that the bond wedging in the 3D structure is inconsistent with the 3D coordinates; it’s just used to indicate that there is an atropisomer bond, the actual stereochemistry of the bond is determined by the 3D coordinates.

_images/atrop_example5.png

Interpreting atropisomers without coordinates

Just as it is possible to interpret atomic and double bond stereochemistry from SMILES without providing atomic coordinates, the RDKit adopts (starting with the 2024.03.2 release) a convention for interpreting bond wedge information in CXSMILES that do not have coordinates provided.

In order to understand how this representation works, a quick digression is necessary to explain the difference in the way bonds are numbered in CXSMILES and the RDKit. In the RDKit ring closure bonds are added last; they are the highest numbered bonds in the molecule. In CXSMILES, the ring closure bonds are added immediately upon closing the ring. Here’s an illustration of this for the SMILES CC1=CC=CC(I)=C1N1C(C)=CC=C1Br; the RDKit bond numbering (what we saw above) is shown first and the CXSMILES numbering is shown second.

_images/atrop_representation2.png

To describe this atropisomer in CXSMILES without using coordinates, we adopt the convention that the atropisomeric bond’s stereochemistry is CCW when the bond to the lowest-numbered neighbor of the start atom (in this case the neighbor connected via bond 7) is wedged and write the CXSMILES for this molecule as CC1=CC=CC(I)=C1N1C(C)=CC=C1Br |wU:7.7||. The stereo values for all possible combinations are shown here:

from

to

bond index

type

stereo

7

1

7

wedge

CCW

wU7.7

7

5

6

wedge

CW

wU7.6

8

9

9

wedge

CW

wU8.9

8

13

14

wedge

CCW

wU8.14

7

1

7

dash

CW

wD7.7

7

5

6

dash

CCW

wD7.6

8

9

9

dash

CCW

wD8.9

8

13

14

dash

CW

wD8.14

It’s also possible to wedge multiple neighbor bonds around an atropisomeric bond, but the wedging must be consistent based on the table above; so the CXSMILES CC1=CC=CC(I)=C1N1C(C)=CC=C1Br |wD:7.7,wU:8.9| is valid, but CC1=CC=CC(I)=C1N1C(C)=CC=C1Br |wD:7.7,wU:8.14| is not.

Validation of Atropisomers

Invalid atropisomers (for example, those with two equivalent groups on one end) will be removed when reading molecules in with sanitization enabled or by calling cleanupAtropisomers(mol).

Searching and Canonicalization

Atropisomers are supported in the canonicalization algorithm of RDKit. They are ignored in substructure searching and similarity searching at this time.

Query Features in Molecule Drawings

Compactly and clearly including information about query features in molecule drawings is a challenging problem. This is definitely a work in progress, but this section describes what is currently supported.

Query Bonds

Here is an example image showing how different bond and query-bond types are rendered.

_images/query_bonds.png

There’s clearly some room for improvement here, for example, it’s not trivial to distinguish “Any” bonds from query bonds where no special handling has been implemented (“other” query types):

_images/query_bonds.2.png

Query Atoms

At the moment the only real support for atomic query features is rendering of atom lists (and “NOT” atom lists); other atomic queries are rendered with a simple ?:

_images/query_atoms.png

Conformer Generation

Introduction

The RDKit can generate conformers for molecules using two different methods. The original method used distance geometry. [16] The default algorithm followed is:

  1. The molecule’s distance bounds matrix is calculated based on the connection table and a set of rules.

  2. The bounds matrix is smoothed using a triangle-bounds smoothing algorithm.

  3. A random distance matrix that satisfies the bounds matrix is generated.

  4. This distance matrix is embedded in 3D dimensions (producing coordinates for each atom).

  5. The resulting coordinates are cleaned up somewhat using the “distance geometry force field”, based on distance constraints from the bounds matrix.

The RDKit also has an implementation of the ETKDG method of Riniker and Landrum [17] which modifies step 5 above to also use torsion angle preferences from the Cambridge Structural Database (CSD) to correct the conformers after distance geometry has been used to generate them. The ETDKDG approach can be extended to include additional torsion terms for small rings and/or macrocycles [18].

When using the ETKDG approaches the quality of the conformers generated is generally good enough to allow them to be used “as is” (i.e. without a subsequent minimization step with another force field) for many applications.

Parameters Controlling Conformer Generation

A large number of parameters which allow control over the conformer generation process are available in the EmbedParameters class. A subset of particularly useful parameters are described here:

  • randomSeed: (default -1) allows you to set a random seed to allow reproducible results

  • numThreads: (default 1) sets the number of compute threads to be used when generating multiple conformers. If set to 0 this will use the maximum number of threads allowed on your system.

  • useRandomCoords: (default False) if set to True then random-coordinate embedding will be done: instead of steps 3. and 4. above, the atoms will be randomly placed in a box and then their positions will be minimized with the “distance geometry force field” in step 5. This approach was described in reference [19]

  • enforceChirality: (default True) ensures that the chirality of specified stereocenters in the molecule is preserved in the conformers.

  • embedFragsSeparately: (default True) for molecules made up of multiple disconnected fragments, this cause conformers of the fragments to be generated independently of each other.

  • coordMap: (default empty) can be used to provide 3D coordinates which will be used to constrain the positions of some of the atoms in the molecule.

  • boundsMat: (default empty) can be used to provide the distance bounds matrix for the molecule.

  • useExpTorsionAnglePrefs: (default False) use the ET part of ETKDG [17]

  • useBasicKnowledge: (default False) use the K part of ETKDG [17]

  • ETVersion: (default 1) specify the version of the standard torsion definitions to use. NOTE for both ETKDGv2 and ETKDGv3 this should be 2 since ETKDGv3 uses the ETKDGv2 definitions for standard torsions (apologies for the confusing numbering)

  • useSmallRingTorsions: (default False) use the sr part of srETDKGv3 [18]

  • useMacrocycleTorsions: (default False) use the macrocycle torsions from ETKDGv3 [18]

  • useMacrocycle14config: (default False) use the 1-4 distance bounds from ETKDGv3 [18]

  • forceTransAmides: (default True) constrain amide bonds to be trans

  • pruneRMsThresh: (default -1.0) if >0.0 this turns on RMSD pruning of the conformers

  • onlyHeavyAtomsForRMS: (default: False) toggles ignoring H atoms when doing RMSD pruning

  • useSymmetryForPruning: (default True) uses symmetry to calculate the minimum RMSD between two conformers when doing RMSD pruning. Note that enabling this causes the RMSD computation to act as if onlyHeavyAtomsForRMS is set to true (even if the parameter itself is set to False).

Note that there are pre-configured parameter objects for the available ETKDG versions: ETKDG, ETKDGv2, ETKDGv3, and srETKDGv3

Additional Information About the Fingerprints

This section, which is not currently comprehensive, is intended to provide some documentation about the types of fingerprints available in the RDKit. We don’t reproduce information that can be found in the literature, but try and capture the unpublished bits (of which there are quite a few).

RDKit Fingerprints

This is an RDKit-specific fingerprint that is inspired by (though it differs significantly from) public descriptions of the Daylight fingerprint [7]. The fingerprinting algorithm identifies all subgraphs in the molecule within a particular range of sizes, hashes each subgraph to generate a raw bit ID, mods that raw bit ID to fit in the assigned fingerprint size, and then sets the corresponding bit. Options are available to generate count-based forms of the fingerprint or “non-folded” forms (using a sparse representation).

The default scheme for hashing subgraphs is to hash the individual bonds based on:
  • the types of the two atoms. Atom types include the atomic number (mod 128), and whether or not the atom is aromatic.

  • the degrees of the two atoms in the path.

  • the bond type (or AROMATIC if the bond is marked as aromatic)

Fingerprint-specific options

  • minPath and maxPath control the size (in bonds) of the subgraphs/paths considered

  • nBitsPerHash: If this is greater than one, each subgraph will set more than one bit. The additional bits will be generated by seeding a random number generator with the original raw bit ID and generating the appropriate number of random numbers.

  • useHs: toggles whether or not Hs are included in the subgraphs/paths (assuming that there are Hs in the molecule graph.

  • tgtDensity: if this is greater than zero, the fingerprint will be repeatedly folded in half until the density of set bits is greater than or equal to this value or the fingerprint only contains minSize bits. Note that this means that the resulting fingerprint will not necessarily be the size you requested.

  • branchedPaths: if this is true (the default value), the algorithm will use subgraphs (i.e features can be branched. If false, only linear paths will be considered.

  • useBondOrder: if true (the default) bond types will be considered when hashing subgraphs, otherwise this component of the hash will be ignored.

Pattern Fingerprints

These fingerprints were designed to be used in substructure screening. These are, as far as I know, unique to the RDKit. The algorithm identifies features in the molecule by doing substructure searches using a small number (12 in the 2019.03 release of the RDKit) of very generic SMARTS patterns - like [*]~[*]~[*](~[*])~[*] or [R]~1[R]~[R]~[R]~1, and then hashing each occurrence of a pattern based on the atom and bond types involved. The fact that particular pattern matched the molecule at all is also stored by hashing the pattern ID and size. If a particular feature contains either a query atom or a query bond (e.g. something generated from SMARTS), the only information that is hashed is the fact that the generic pattern matched.

For the 2019.03 release, the atom types use just the atomic number of the atom and the bond types use the bond type, or AROMATIC for aromatic bonds).

NOTE: Because it plays an important role in substructure screenout, the internals of this fingerprint (the generic patterns used and/or the details of the hashing algorithm) may change from one release to the next.

Atom-Pair and Topological Torsion Fingerprints

These two related fingerprints are implemented based on the original papers: [8] [9]. Atoms are typed based on atomic number, number of pi electrons, and the degree of the atom. Optionally information about atomic chirality can also be integrated into the atom types. Both fingerprint types can be generated in explicit or sparse form and as bit or count vectors. These fingerprint types are different from the others in the RDKit in that bits in the sparse form of the fingerprint can be directly explained (i.e. the “hashing function” used is fully reversible).

These fingerprints were originally “intended” to be used in count-vectors and they seem to work better that way. The default behavior of the explicit bit-vector forms of both fingerprints is to use a “count simulation” procedure where multiple bits are set for a given feature if it occurs more than once. The default behavior is to use 4 fingerprint bits for each feature (so a 2048 bit fingerprint actually stores information about the same number of features as a 512 bit fingerprint that isn’t using count simulation). The bins correspond to counts of 1, 2, 4, and 8. As an example of how this works: if a feature occurs 5 times in a molecule, the bits corresponding to counts 1, 2, and 4 will be set.

Morgan and Feature Morgan Fingerprints

These are implemented based on the original paper [10]. The algorithm follows the description in the paper as closely as possible with the exception of the chemical feature definitions used for the “Feature Morgan” fingerprint - the RDKit implementation uses the feature types Donor, Acceptor, Aromatic, Halogen, Basic, and Acidic with definitions adapted from those in the paper [11]. It is possible to provide your own atom types. The fingerprints are available as either explicit or sparse count vectors or explicit bit vectors.

Layered Fingerprints

These are another “RDKit original” and were developed with the intention of using them as a substructure fingerprint. Since the pattern fingerprint is far simpler and has proven to be quite effective as a substructure fingerprint, the layered fingerprint hasn’t received much attention. It may still be interesting for something, so we continue to include it.

The idea of the fingerprint is generate features using the same subgraph (or path) enumeration algorithm used in the RDKit fingerprint. After a subgraph has been generated, it is used to set multiple bits based on different atom and bond type definitions.

Feature Flags: global variables affecting RDKit behavior

The RDKit uses a number of “feature flags”: global variables which affect its behavior. These have generally been added to maintain backwards compatibility when introducing new algorithms which yield different results.

Here’s are the current feature flags:

  • preferCoordGen: when this is true Schrodinger’s open-source Coordgen library will be used to generate 2D coordinates of molecules. The default value is false. This can be set from C++ using the variable RDKit::RDDepict::preferCoordGen or from Python using the function rdDepictor.SetPreferCoordGen(). Added in the 2018.03 release.

  • allowNontetrahedralChirality: when this is true non-tetrahedral chirality will be perceived from 3D coordinates. The default value is true unless the environment variable RDK_ENABLE_NONTETRAHEDRAL_STEREO is set to "0". Can set/checked from C++ using the functions RDKit::Chirality::setAllowNontetrahedralChirality() / RDKit::Chirality::getAllowNontetrahedralChirality() or from Python using the functions Chem.SetAllowNontetrahedralChirality() / Chem.GetAllowNontetrahedralChirality(). Added in the 2022.09 release.

  • useLegacyStereoPerception: when this is true the legacy implementation for perceiving stereochemistry will be used. The default value is true unless the environment variable RDK_USE_LEGACY_STEREO_PERCEPTION is set to "0". Can set/checked from C++ using the functions RDKit::Chirality::setUseLegacyStereoPerception() / RDKit::Chirality::getUseLegacyStereoPerception() or from Python using the functions Chem.SetUseLegacyStereoPerception() / Chem.GetUseLegacyStereoPerception(). Added in the 2022.09 release.

Footnotes

License

_images/picture_5.png

This document is copyright (C) 2007-2021 by Greg Landrum

This work is licensed under the Creative Commons Attribution-ShareAlike 4.0 License. To view a copy of this license, visit http://creativecommons.org/licenses/by-sa/4.0/ or send a letter to Creative Commons, 543 Howard Street, 5th Floor, San Francisco, California, 94105, USA.

The intent of this license is similar to that of the RDKit itself. In simple words: “Do whatever you want with it, but please give us some credit.”