Conformer generation¶

Conformer generation is based on four-spatial dimension Distance Geometry. This library’s implementation features the following:

A spatial model generates atom-pairwise bounds on their distance in the final conformations and four-atom chiral constraints when distance bounds cannot encompass chiral elements of complex shapes. For large shapes, chiral information is captured by using multiple chiral constraints.
The distance bounds are smoothed to conform to the triangle inequalities. After each successive choice of a fixed distance between the bounds, you can choose to re-smooth all bounds (full metrization) or stop re-smoothing after a fixed number of chosen distances (partial metrization).
The bounds are embedded in four dimensions and refined in three stages, permitting the chiral constraints to invert by expanding into four dimensions, and then compressing the fourth dimension back out. Lastly, dihedral error terms are minimized.
The refinement error function is modified to enable the placement of haptic ligand’s bonding atoms’ average position at shapes’ idealized ligand sites.

class scine_molassembler.dg.Configuration¶

A configuration object for distance geometry runs with sane defaults

property fixed_positions¶

Set fixed positions for a subset of atoms in bohr units. Defaults to no fixed positions.

Any fixed atom must have zero, one or all binding sites fully fixed. No individual sites may be partially fixed (i.e. the atoms constituting a haptic ligand binding site must be either completely free or completely fixed and nothing in between).

property partiality¶: Choose for how many atoms to re-smooth the distance bounds after a distance choice. Defaults to four-atom partiality.

property refinement_gradient_target¶: Sets the gradient at which a refinement is considered complete. Defaults to 1e-5.

property refinement_step_limit¶: Sets the maximum number of refinement steps. Defaults to 10’000.

property spatial_model_loosening¶: Set loosening factor for spatial model (1.0 is no loosening, 2.0 is strong loosening). Defaults to 1.0.

class scine_molassembler.dg.Error¶

Things that can go wrong in Distance Geometry

Members:

ZeroAssignmentStereopermutators :
The molecule you are trying to generate conformers for has zero-assignment stereopermutators, meaning that it is not representable in three dimensional space.

AtomStereopermutators remove those stereopermutations from the user-accessible set that it deems obviously impossible. This includes overlapping haptic ligand binding cones and multidentate ligand bridges that are too short to span the angle needed in a stereopermutation. In most cases, this will simply eliminate trans-arranged multidentate ligands with too short bridges. It is conservative, however, and may not be strict enough to eliminate all stereopermutators with bridges that are too short to comply with the spatial modeling. In that case, you may get GraphImpossible.

If you get this error, reconsider whether your input graph is reasonable and representable in three dimensions. If you believe the atom stereopermutator incorrectly has zero assignments, please contact us and open an issue.

GraphImpossible :
The molecule you are trying to generate conformers for is either incompatible with the applied spatial model or plain not representable in three dimensions.

The applied spatial model is not very smart and mostly applies simple geometric considerations. One one hand, it may be that centers whose shapes are heavily distorted due to e.g. multiple small cycles are not recognized correctly or modeled loosely enough in order for a conformer to be possible. On the other hand, it is also possible to create graphs that are not representable in three dimensions. In both circumstances, you will get this error to indicate that the spatial model cannot deal with your input graph.

If you get this error, reconsider whether your input graph is reasonable and representable in three dimensions. If you are sure it is, please contact us and open an issue.

RefinementException :
An exception occurred during refinement

The form of the potential during Distance Geometry refinement can be exceptional due to e.g. divisions by zero. These exceptions can occur completely randomly and have no bearing on validity of input.

If you get this error, generate some more conformers. If all of your inputs yield refinement exceptions, there might be a modeling problem, so please contact us and open an issue.

RefinementMaxIterationsReached :
Refinement could not find a minimum in your specified maximum number of iterations

Typically, this may mean that the Molecule you are trying to generate conformers is either way too big for the number of iterations in the potential minimization or that refinement got stuck in some sort of bad back and forwards.

Try adjusting your number of iterations. If the problem persists, there may be a problem with the form of the refinement potential, so please contact us and open an issue.

RefinedStructureInacceptable :
The result of a refinement did not meet criteria for acceptance

In Distance Geometry, we generate a list of atom-pairwise distance bounds that indicate the minimum and maximum distance that atom pair should have in a final conformation. Additionally, chiral constraints are generated (similar to improper dihedrals) that indicate whether a chiral element is arranged correctly. Refinement, which tries to minimize these errors, may end up in a local minimum that still violates some of these bounds.

This is purely a stochastic problem and should not reflect on your inputs. If you get this error, generate some more conformers. If all of your inputs yield this error, there may be a problem with the refinement potential, so please contact us and open an issue.

RefinedChiralsWrong :
Chiral constraints on the refined structure are still incorrect

After refinement, chiral constraints are completely wrong. This is a rare stochastic error and should not reflect on your inputs.

Generate more conformers. If you get this error a lot on your inputs, there may be a problem with the refinement potential, so please contact us and open an issue.

DecisionListMismatch :
In directed conformer generation, failed to generate decision list

UnknownException : Unknown exception occurred. Please report this as an issue to the developers!

DecisionListMismatch = Error.DecisionListMismatch¶

GraphImpossible = Error.GraphImpossible¶

RefinedChiralsWrong = Error.RefinedChiralsWrong¶

RefinedStructureInacceptable = Error.RefinedStructureInacceptable¶

RefinementException = Error.RefinementException¶

RefinementMaxIterationsReached = Error.RefinementMaxIterationsReached¶

UnknownException = Error.UnknownException¶

ZeroAssignmentStereopermutators = Error.ZeroAssignmentStereopermutators¶

property name¶

handle) -> str

Type: (self

class scine_molassembler.dg.Partiality¶

Limit triangle inequality bounds smoothing to a subset of all atoms

Members:

FourAtom : Resmooth only after each of the first four atom choices

TenPercent : Resmooth for the first 10% of all atoms

All : Resmooth after each distance choice

All = Partiality.All¶

FourAtom = Partiality.FourAtom¶

TenPercent = Partiality.TenPercent¶

property name¶

handle) -> str

Type: (self

scine_molassembler.dg.generate_conformation(molecule: scine_molassembler.Molecule, seed: int, configuration: scine_molassembler.dg.Configuration = (partiality=Partiality.FourAtom, refinement_step_limit=10000, refinement_gradient_target=1e-05, spatial_model_loosening=1.0, fixed_positions=[])) → Union[numpy.ndarray[float64[m, 3]], scine_molassembler.dg.Error]¶

Generate 3D positions for a molecule.

In the case of a molecule that does not have unassigned stereopermutators, this is akin to generating a conformer. If there are unassigned stereopermutators, these are assigned at random (consistent with relative statistical occurrences of stereopermutations) for each structure. If, for instance, your molecules contains a single unassigned asymmetric tetrahedron atom stereopermutator, the resulting conformation will be one of both assignments.

Parameters

molecule – Molecule to generate positions for. May not contain stereopermutators with zero assignments (no feasible stereopermutations).
seed – Seed with which to initialize a PRNG with for the conformer generation procedure.
configuration – Detailed Distance Geometry settings. Defaults are usually fine.

Return type

Either a position result or an error string explaining why conformer generation failed.

>>> # Generate a single conformation
>>> mol = io.experimental.from_smiles("N[C@](Br)(O)F")
>>> conformation = generate_conformation(mol, 110)
>>> isinstance(conformation, Error) # Did the conformer generation fail?
False
>>> type(conformation) # Successful results have matrix type:
<class 'numpy.ndarray'>

scine_molassembler.dg.generate_ensemble(molecule: scine_molassembler.Molecule, num_structures: int, seed: int, configuration: scine_molassembler.dg.Configuration = (partiality=Partiality.FourAtom, refinement_step_limit=10000, refinement_gradient_target=1e-05, spatial_model_loosening=1.0, fixed_positions=[])) → List[Union[numpy.ndarray[float64[m, 3]], scine_molassembler.dg.Error]]¶

Generate a set of 3D positions for a molecule.

In the case of a molecule that does not have unassigned stereopermutators, this is akin to generating a conformational ensemble. If there are unassigned stereopermutators, these are assigned at random (consistent with relative statistical occurrences of stereopermutations) for each structure. If, for instance, your molecules contains a single unassigned asymmetric tetrahedron atom stereopermutator, the ensemble will contain confomers of both assignments, akin to a racemic mixture.

Note

This function is parallelized and will utilize OMP_NUM_THREADS threads. The resulting list is sequenced and reproducible given the same seed.

Parameters

molecule – Molecule to generate positions for. May not contain stereopermutators with zero assignments (no feasible stereopermutations).
num_structures – Number of desired structures to generate
configuration – Detailed Distance Geometry settings. Defaults are usually fine.

Return type

Heterogeneous list of either a position result or an error string explaining why conformer generation failed.

>>> # Generate a conformational ensemble
>>> butane = io.experimental.from_smiles("CCCC")
>>> seed = 1010
>>> results = generate_ensemble(butane, 10, seed)
>>> # Each element in the list can be either a string or a positions matrix
>>> # So let's see how many failed:
>>> sum([1 if isinstance(r, Error) else 0 for r in results])
0

scine_molassembler.dg.generate_random_conformation(molecule: scine_molassembler.Molecule, configuration: scine_molassembler.dg.Configuration = (partiality=Partiality.FourAtom, refinement_step_limit=10000, refinement_gradient_target=1e-05, spatial_model_loosening=1.0, fixed_positions=[])) → Union[numpy.ndarray[float64[m, 3]], scine_molassembler.dg.Error]¶

Generate 3D positions for a molecule.

In the case of a molecule that does not have unassigned stereopermutators, this is akin to generating a conformer. If there are unassigned stereopermutators, these are assigned at random (consistent with relative statistical occurrences of stereopermutations) for each structure. If, for instance, your molecules contains a single unassigned asymmetric tetrahedron atom stereopermutator, the resulting conformation will be one of both assignments.

Parameters

molecule – Molecule to generate positions for. May not contain stereopermutators with zero assignments (no feasible stereopermutations).
configuration – Detailed Distance Geometry settings. Defaults are usually fine.

Return type

Either a position result or an error string explaining why conformer generation failed.

Note

This function advances molassembler’s global PRNG state.

>>> # Generate a single conformation
>>> mol = io.experimental.from_smiles("N[C@](Br)(O)F")
>>> conformation = generate_random_conformation(mol)
>>> isinstance(conformation, Error) # Did the conformer generation fail?
False
>>> type(conformation) # Successful results have matrix type:
<class 'numpy.ndarray'>

scine_molassembler.dg.generate_random_ensemble(molecule: scine_molassembler.Molecule, num_structures: int, configuration: scine_molassembler.dg.Configuration = (partiality=Partiality.FourAtom, refinement_step_limit=10000, refinement_gradient_target=1e-05, spatial_model_loosening=1.0, fixed_positions=[])) → List[Union[numpy.ndarray[float64[m, 3]], scine_molassembler.dg.Error]]¶

Generate a set of 3D positions for a molecule.

In the case of a molecule that does not have unassigned stereopermutators, this is akin to generating a conformational ensemble. If there are unassigned stereopermutators, these are assigned at random (consistent with relative statistical occurrences of stereopermutations) for each structure. If, for instance, your molecules contains a single unassigned asymmetric tetrahedron atom stereopermutator, the ensemble will contain confomers of both assignments, akin to a racemic mixture.

Note

This function is parallelized and will utilize OMP_NUM_THREADS threads. The resulting list is sequenced and reproducible given the same global PRNG state.

Note

This function advances molassembler’s global PRNG state.

Parameters

molecule – Molecule to generate positions for. May not contain stereopermutators with zero assignments (no feasible stereopermutations).
num_structures – Number of desired structures to generate
configuration – Detailed Distance Geometry settings. Defaults are usually fine.

Return type

Heterogeneous list of either a position result or an error string explaining why conformer generation failed.

>>> # Generate a conformational ensemble
>>> butane = io.experimental.from_smiles("CCCC")
>>> results = generate_random_ensemble(butane, 10)
>>> # Each element in the list can be either a string or a positions matrix
>>> # So let's see how many failed:
>>> sum([1 if isinstance(r, Error) else 0 for r in results])
0