#!/usr/local/bin/env python
# ==============================================================================
# FILE DOCSTRING
# ==============================================================================
"""
Restraints
==========
Automated selection and imposition of receptor-ligand restraints for absolute
alchemical binding free energy calculations, along with computation of the
standard state correction.
"""
# ==============================================================================
# GLOBAL IMPORTS
# ==============================================================================
import abc
import math
import random
import inspect
import logging
import functools
import itertools
import numpy as np
import scipy.integrate
import mdtraj as md
import openmmtools as mmtools
from simtk import openmm, unit
from . import pipeline
from .utils import methoddispatch
logger = logging.getLogger(__name__)
# ==============================================================================
# MODULE CONSTANTS
# ==============================================================================
V0 = 1660.53928 * unit.angstroms**3 # standard state volume
# ==============================================================================
# CUSTOM EXCEPTIONS
# ==============================================================================
[docs]class RestraintStateError(mmtools.states.ComposableStateError):
"""Error raised by an :class:`RestraintState`."""
pass
[docs]class RestraintParameterError(Exception):
"""Error raised by a :class:`ReceptorLigandRestraint`."""
pass
# ==============================================================================
# Dispatch appropriate restraint type from registered restraint classes
# ==============================================================================
[docs]def available_restraint_classes():
"""
Return all available restraint classes.
Returns
-------
restraint_classes : dict of {str : class}
``restraint_classes[name]`` is the class corresponding to ``name``
"""
# Get a list of all subclasses of ReceptorLigandRestraint
def get_all_subclasses(check_cls):
"""Find all subclasses of a given class recursively."""
all_subclasses = []
for subclass in check_cls.__subclasses__():
all_subclasses.append(subclass)
all_subclasses.extend(get_all_subclasses(subclass))
return all_subclasses
# Build an index of all names, ensuring there are no name collisions.
available_restraints = dict()
for cls in get_all_subclasses(ReceptorLigandRestraint):
classname = cls.__name__
if inspect.isabstract(cls):
# Skip abstract base classes
pass
elif classname in available_restraints:
raise ValueError("More than one restraint subclass has the name '{}'.".format(classname))
else:
available_restraints[classname] = cls
return available_restraints
[docs]def available_restraint_types():
"""
List all available restraint types.
Returns
-------
available_restraint_types : list of str
List of names of available restraint classes
"""
available_restraints = available_restraint_classes()
return available_restraints.keys()
[docs]def create_restraint(restraint_type, **kwargs):
"""Factory of receptor-ligand restraint objects.
Parameters
----------
restraint_type : str
Restraint type name matching a register (imported) subclass of :class:`ReceptorLigandRestraint`.
kwargs
Parameters to pass to the restraint constructor.
"""
available_restraints = available_restraint_classes()
if restraint_type not in available_restraints:
raise ValueError("Restraint type {} unknown. Options are: {}".format(
restraint_type, str(available_restraints.keys())))
cls = available_restraints[restraint_type]
return cls(**kwargs)
# ==============================================================================
# ComposableState class to control the strength of restraints.
# ==============================================================================
[docs]class RestraintState(object):
"""
The state of a restraint.
A ``ComposableState`` controlling the strength of a restraint
through its ``lambda_restraints`` property.
Parameters
----------
lambda_restraints : float
The strength of the restraint. Must be between 0 and 1.
Attributes
----------
lambda_restraints
Examples
--------
Create a system in a thermodynamic state.
>>> from openmmtools import testsystems, states
>>> system_container = testsystems.LysozymeImplicit()
>>> system, positions = system_container.system, system_container.positions
>>> thermodynamic_state = states.ThermodynamicState(system, 300*unit.kelvin)
>>> sampler_state = states.SamplerState(positions)
Identify ligand atoms. Topography automatically identify receptor atoms too.
>>> from yank.yank import Topography
>>> topography = Topography(system_container.topology, ligand_atoms=range(2603, 2621))
Apply a Harmonic restraint between receptor and protein. Let the restraint
automatically determine all the parameters.
>>> restraint = Harmonic()
>>> restraint.determine_missing_parameters(thermodynamic_state, sampler_state, topography)
>>> restraint.restrain_state(thermodynamic_state)
Create a ``RestraintState`` object to control the strength of the restraint.
>>> restraint_state = RestraintState(lambda_restraints=1.0)
``RestraintState`` implements the ``IComposableState`` interface, so it can be
used with ``CompoundThermodynamicState``.
>>> compound_state = states.CompoundThermodynamicState(thermodynamic_state=thermodynamic_state,
... composable_states=[restraint_state])
>>> compound_state.lambda_restraints
1.0
>>> integrator = openmm.VerletIntegrator(1.0*unit.femtosecond)
>>> context = compound_state.create_context(integrator)
>>> context.getParameter('lambda_restraints')
1.0
You can control the parameters in the OpenMM Context by setting the state's
attributes. To To deactivate the restraint, set `lambda_restraints` to 0.0.
>>> compound_state.lambda_restraints = 0.0
>>> compound_state.apply_to_context(context)
>>> context.getParameter('lambda_restraints')
0.0
"""
def __init__(self, lambda_restraints):
self.lambda_restraints = lambda_restraints
@property
def lambda_restraints(self):
"""Float: the strength of the applied restraint (between 0 and 1 inclusive)."""
return self._lambda_restraints
@lambda_restraints.setter
def lambda_restraints(self, value):
assert 0.0 <= value <= 1.0
self._lambda_restraints = float(value)
[docs] def apply_to_system(self, system):
"""
Set the strength of the system's restraint to this.
System is updated in-place
Parameters
----------
system : simtk.openmm.System
The system to modify.
Raises
------
RestraintStateError
If the system does not have any ``CustomForce`` with a
``lambda_restraint`` global parameter.
"""
# Set lambda_restraints in all forces that have it.
for force, parameter_id in self._get_system_forces_parameters(system):
force.setGlobalParameterDefaultValue(parameter_id, self._lambda_restraints)
[docs] def check_system_consistency(self, system):
"""
Check if the system's restraint is in this restraint state.
It raises a :class:`RestraintStateError` if the restraint is not consistent
with the state.
Parameters
----------
system : simtk.openmm.System
The system with the restraint to test.
Raises
------
RestraintStateError
If the system is not consistent with this state.
"""
# Set lambda_restraints in all forces that have it.
for force, parameter_id in self._get_system_forces_parameters(system):
force_lambda = force.getGlobalParameterDefaultValue(parameter_id)
if force_lambda != self.lambda_restraints:
err_msg = 'Consistency check failed: system {}, state {}'
raise RestraintStateError(err_msg.format(force_lambda, self._lambda_restraints))
[docs] def apply_to_context(self, context):
"""Put the restraint in the `Context` into this state.
Parameters
----------
context : simtk.openmm.Context
The context to set.
Raises
------
RestraintStateError
If the context does not have the required lambda global variables.
"""
try:
context.setParameter('lambda_restraints', self._lambda_restraints)
except Exception:
raise RestraintStateError('The context does not have a restraint.')
@classmethod
def _standardize_system(cls, system):
"""Standardize the given system.
Set lambda_restraints of the system to 1.0.
Parameters
----------
system : simtk.openmm.System
The system to standardize.
Raises
------
RestraintStateError
If the system is not consistent with this state.
"""
# Set lambda_restraints to 1.0 in all forces that have it.
for force, parameter_id in cls._get_system_forces_parameters(system):
force.setGlobalParameterDefaultValue(parameter_id, 1.0)
@staticmethod
def _get_system_forces_parameters(system):
"""Yields the system's forces having a ``lambda_restraints`` parameter.
Yields
------
A tuple force, ``parameter_index`` for each force with ``lambda_restraints``.
"""
found_restraint = False
# Retrieve all the forces with global supported parameters.
for force_index in range(system.getNumForces()):
force = system.getForce(force_index)
try:
n_global_parameters = force.getNumGlobalParameters()
except AttributeError:
continue
for parameter_id in range(n_global_parameters):
parameter_name = force.getGlobalParameterName(parameter_id)
if parameter_name == 'lambda_restraints':
found_restraint = True
yield force, parameter_id
# Raise error if the system doesn't have a restraint.
if found_restraint is False:
raise RestraintStateError('The system does not have a restraint.')
def __getstate__(self):
return dict(lambda_restraints=self._lambda_restraints)
def __setstate__(self, serialization):
self.lambda_restraints = serialization['lambda_restraints']
# ==============================================================================
# Base class for receptor-ligand restraints.
# ==============================================================================
ABC = abc.ABCMeta('ABC', (object,), {}) # compatible with Python 2 *and* 3
[docs]class ReceptorLigandRestraint(ABC):
"""
A restraint preventing a ligand from drifting too far from its receptor.
With replica exchange simulations, keeping the ligand close to the binding
pocket can enhance mixing between the interacting and the decoupled state.
This should be always used in implicit simulation, where there are no periodic
boundary conditions.
This restraint strength is controlled by a global context parameter called
``lambda_restraints``. You can easily control this variable through the
``RestraintState`` object.
Notes
-----
Creating a subclass requires the following:
1. Implement a constructor. Optionally this can leave all or a subset of
the restraint parameters undefined. In this case, you need to provide
an implementation of :func:`determine_missing_parameters`.
2. Implement :func:`restrain_state` that add the restrain ``Force`` to the state's
`System`.
3. Implement :func:`get_standard_state_correction` to return standard state correction.
4. Optionally, implement :func:`determine_missing_parameters` to fill in
the parameters left undefined in the constructor.
"""
@abc.abstractmethod
[docs] def restrain_state(self, thermodynamic_state):
"""Add the restraint force to the state's `System`.
Parameters
----------
thermodynamic_state : openmmtools.states.ThermodynamicState
The thermodynamic state holding the system to modify.
"""
pass
@abc.abstractmethod
[docs] def get_standard_state_correction(self, thermodynamic_state):
"""Return the standard state correction.
Parameters
----------
thermodynamic_state : openmmtools.states.ThermodynamicState
The thermodynamic state.
"""
pass
[docs] def determine_missing_parameters(self, thermodynamic_state, sampler_state, topography):
"""
Automatically choose undefined parameters.
Optionally, a :class:`ReceptorLigandRestraint` can support the automatic
determination of all or a subset of the parameters that can be
left undefined in the constructor, making implementation of this method optional.
Parameters
----------
thermodynamic_state : openmmtools.states.ThermodynamicState
The thermodynmaic state to inspect
sampler_state : openmmtools.states.SamplerState
The sampler state holding the positions of all atoms.
topography : yank.Topography
The topography with labeled receptor and ligand atoms.
"""
raise NotImplementedError('{} does not support automatic determination of the '
'restraint parameters'.format(self.__class__.__name__))
class _RestrainedAtomsProperty(object):
"""
Descriptor of restrained atoms.
Casts generic iterables of ints into lists.
"""
def __init__(self, atoms_type):
self._atoms_type = atoms_type
@property
def _attribute_name(self):
"""Name of the internally stored variable (read-only)."""
return '_restrained_' + self._atoms_type + '_atoms'
def __get__(self, instance, owner_class=None):
return getattr(instance, self._attribute_name)
def __set__(self, instance, new_restrained_atoms):
# If we set the restrained attributes to None, no reason to check things.
if new_restrained_atoms is not None:
new_restrained_atoms = self._validate_atoms(new_restrained_atoms)
setattr(instance, self._attribute_name, new_restrained_atoms)
@methoddispatch
def _validate_atoms(self, restrained_atoms):
"""Casts a generic iterable of ints into a list to support concatenation."""
try:
restrained_atoms = restrained_atoms.tolist()
except AttributeError:
restrained_atoms = list(restrained_atoms)
return restrained_atoms
# ==============================================================================
# Base class for radially-symmetric receptor-ligand restraints.
# ==============================================================================
[docs]class RadiallySymmetricRestraint(ReceptorLigandRestraint):
"""
Base class for radially-symmetric restraints between ligand and protein.
The restraint is applied between the centroids of two groups of atoms
that belong to the receptor and the ligand respectively. The centroids
are determined by a mass-weighted average of the group particles positions.
The restraint strength is controlled by a global context parameter called
'lambda_restraints'.
With OpenCL, groups with more than 1 atom are supported only on 64bit
platforms.
The class allows the restrained atoms to be temporarily undefined, but in
this case, :func:`determine_missing_parameters` must be called before using
the restraint.
Parameters
----------
restrained_receptor_atoms : iterable of int, int, or str, optional
The indices of the receptor atoms to restrain, an MDTraj DSL expression, any other
:class:`Topography <yank.Topography>` region name,
or :func:`Topography Selection <yank.Topography.select>`.
This can temporarily be left undefined, but :func:`determine_missing_parameters`
must be called before using the Restraint object. The same if a DSL
expression or Topography selection is provided (default is None).
restrained_ligand_atoms : iterable of int, int, or str, optional
The indices of the ligand atoms to restrain, an MDTraj DSL expression, or a
:class:`Topography <yank.Topography>` region name,
or :func:`Topography Selection <yank.Topography.select>`.
This can temporarily be left undefined, but :func:`determine_missing_parameters`
must be called before using the Restraint object. The same if a DSL
expression or Topography selection is provided (default is None).
Attributes
----------
restrained_receptor_atoms : list of int, str, or None
The indices of the receptor atoms to restrain, an MDTraj selection string, or a Topography selection
string.
restrained_ligand_atoms : list of int, str, None
The indices of the receptor atoms to restrain, an MDTraj selection string, or a Topography selection
string.
Notes
-----
To create a subclass, follow these steps:
1. Implement the property :func:`_energy_function` with the energy function of choice.
2. Implement the property :func:`_bond_parameters` to return the :func:`_energy_function`
parameters as a dict ``{parameter_name: parameter_value}``.
3. Optionally, you can overwrite the :func:`_determine_bond_parameters` member
function to automatically determine these parameters from the atoms positions.
"""
def __init__(self, restrained_receptor_atoms=None, restrained_ligand_atoms=None):
self.restrained_receptor_atoms = restrained_receptor_atoms
self.restrained_ligand_atoms = restrained_ligand_atoms
# -------------------------------------------------------------------------
# Public properties.
# -------------------------------------------------------------------------
class _RadiallySymmetricRestrainedAtomsProperty(_RestrainedAtomsProperty):
"""
Descriptor of restrained atoms.
Extends `_RestrainedAtomsProperty` to handle single integers and strings.
"""
_CENTROID_COMPUTE_STRING = ("You are specifying {} {} atoms, "
"the final atoms will be chosen as the centroid of this set.")
@methoddispatch
def _validate_atoms(self, restrained_atoms):
restrained_atoms = super()._validate_atoms(restrained_atoms)
if len(restrained_atoms) > 1:
logger.debug(self._CENTROID_COMPUTE_STRING.format("more than one", self._atoms_type))
return restrained_atoms
@_validate_atoms.register(str)
def _validate_atoms_string(self, restrained_atoms):
warn_string = self._CENTROID_COMPUTE_STRING.format("a string for", self._atoms_type)
warn_string += 'but you MUST run "determine_missing_parameters" to process the string'
logger.warning(warn_string)
return restrained_atoms
@_validate_atoms.register(int)
def _validate_atoms_int(self, restrained_atoms):
return [restrained_atoms]
restrained_receptor_atoms = _RadiallySymmetricRestrainedAtomsProperty('receptor')
restrained_ligand_atoms = _RadiallySymmetricRestrainedAtomsProperty('ligand')
# -------------------------------------------------------------------------
# Public methods.
# -------------------------------------------------------------------------
[docs] def restrain_state(self, thermodynamic_state):
"""Add the restraining Force(s) to the thermodynamic state's system.
All the parameters must be defined at this point. An exception is
raised if they are not.
Parameters
----------
thermodynamic_state : openmmtools.states.ThermodynamicState
The thermodynamic state holding the system to modify.
Raises
------
RestraintParameterError
If the restraint has undefined parameters.
"""
# Check that restrained atoms are defined.
if not self._are_restrained_atoms_defined:
raise RestraintParameterError('Restraint {}: Undefined restrained '
'atoms.'.format(self.__class__.__name__))
# Create restraint force.
restraint_force = self._create_restraint_force(self.restrained_receptor_atoms,
self.restrained_ligand_atoms)
# Set periodic conditions on the force if necessary.
restraint_force.setUsesPeriodicBoundaryConditions(thermodynamic_state.is_periodic)
# Get a copy of the system of the ThermodynamicState, modify it and set it back.
system = thermodynamic_state.system
system.addForce(restraint_force)
thermodynamic_state.system = system
[docs] def get_standard_state_correction(self, thermodynamic_state):
"""Return the standard state correction.
Parameters
----------
thermodynamic_state : openmmtools.states.ThermodynamicState
The thermodynamic state.
Returns
-------
correction : float
The unit-less standard-state correction, in kT (at the
temperature of the given thermodynamic state).
"""
benchmark_id = 'Restraint {}: Computing standard state correction'.format(self.__class__.__name__)
timer = mmtools.utils.Timer()
timer.start(benchmark_id)
r_min = 0 * unit.nanometers
r_max = 100 * unit.nanometers # TODO: Use maximum distance between atoms?
# Create a System object containing two particles connected by the reference force
system = openmm.System()
system.addParticle(1.0 * unit.amu)
system.addParticle(1.0 * unit.amu)
force = self._create_restraint_force([0], [1])
# Disable the PBC if on for this approximation of the analytical solution
force.setUsesPeriodicBoundaryConditions(False)
system.addForce(force)
# Create a Reference context to evaluate energies on the CPU.
integrator = openmm.VerletIntegrator(1.0 * unit.femtoseconds)
platform = openmm.Platform.getPlatformByName('Reference')
context = openmm.Context(system, integrator, platform)
# Set default positions.
positions = unit.Quantity(np.zeros([2,3]), unit.nanometers)
context.setPositions(positions)
# Create a function to compute integrand as a function of interparticle separation.
beta = thermodynamic_state.beta
def integrand(r):
"""
Parameters
----------
r : float
Inter-particle separation in nanometers
Returns
-------
dI : float
Contribution to integrand (in nm^2).
"""
positions[1, 0] = r * unit.nanometers
context.setPositions(positions)
state = context.getState(getEnergy=True)
potential = state.getPotentialEnergy()
dI = 4.0 * math.pi * r**2 * math.exp(-beta * potential)
return dI
# Integrate shell volume.
shell_volume, shell_volume_error = scipy.integrate.quad(lambda r: integrand(r), r_min / unit.nanometers,
r_max / unit.nanometers) * unit.nanometers**3
logger.debug("shell_volume = %f nm^3" % (shell_volume / unit.nanometers**3))
# The restraint shell volume must be smaller than the
# system box volume or the restraint doesn't make sense.
if thermodynamic_state.is_periodic:
# Compute system volume in NVT/NPT ensemble.
system_box_volume = thermodynamic_state.volume
if system_box_volume is None: # NPT ensemble
box_vectors = thermodynamic_state.system.getDefaultPeriodicBoxVectors()
system_box_volume = mmtools.states._box_vectors_volume(box_vectors)
logger.debug("System volume = {} nm^3".format(system_box_volume / unit.nanometers**3))
# Raise error if shell volume is too big.
if shell_volume > system_box_volume:
raise RuntimeError('The restraint does not limit the configurational '
'space within the solvation box.')
# Compute standard-state volume for a single molecule in a box of
# size (1 L) / (avogadros number). Should also generate constant V0.
liter = 1000.0 * unit.centimeters**3 # one liter
standard_state_volume = liter / (unit.AVOGADRO_CONSTANT_NA*unit.mole) # standard state volume
logger.debug("Standard state volume = {} nm^3".format(standard_state_volume / unit.nanometers**3))
# Compute standard state correction for releasing shell restraints into standard-state box (in units of kT).
DeltaG = - math.log(standard_state_volume / shell_volume)
logger.debug('Standard state correction: {:.3f} kT'.format(DeltaG))
# Report elapsed time.
timer.stop(benchmark_id)
timer.report_timing()
# Return standard state correction (in kT).
return DeltaG
[docs] def determine_missing_parameters(self, thermodynamic_state, sampler_state, topography):
"""Automatically determine missing parameters.
If some parameters have been left undefined (i.e. the atoms to restrain
or the restraint force parameters) this attempts to find them using the
information in the states and the topography.
Parameters
----------
thermodynamic_state : openmmtools.states.ThermodynamicState
The thermodynamic state.
sampler_state : openmmtools.states.SamplerState, optional
The sampler state holding the positions of all atoms.
topography : yank.Topography, optional
The topography with labeled receptor and ligand atoms.
"""
# Determine restrained atoms, if needed.
self._determine_restrained_atoms(sampler_state, topography)
# Determine missing parameters. This is implemented in the subclass.
self._determine_bond_parameters(thermodynamic_state, sampler_state, topography)
# -------------------------------------------------------------------------
# Internal-usage: properties and methods for subclasses.
# -------------------------------------------------------------------------
@abc.abstractproperty
def _energy_function(self):
"""str: energy expression of the restraint force.
This must be implemented by the inheriting class.
"""
pass
@abc.abstractproperty
def _bond_parameters(self):
"""dict: the bond parameters of the restraint force.
This is a dictionary parameter_name: parameter_value that
will be used to configure the `CustomBondForce` added to
the `System`.
If there are parameters undefined, this must be None.
"""
pass
def _determine_bond_parameters(self, thermodynamic_state, sampler_state, topography):
"""Determine the missing bond parameters.
Optionally, a subclass can implement this method to automatically
define the bond parameters of the restraints from the information
in the given states and topography. The default implementation just
raises a NotImplemented error if `_bond_parameters` are undefined.
Parameters
----------
thermodynamic_state : openmmtools.states.ThermodynamicState
The thermodynamic state.
sampler_state : openmmtools.states.SamplerState
The sampler state holding the positions of all atoms.
topography : yank.Topography
The topography with labeled receptor, ligand atoms, and any regions defined.
"""
# Raise exception only if the subclass doesn't already defines parameters.
if self._bond_parameters is None:
raise NotImplementedError('Restraint {} cannot automatically determine '
'bond parameters.'.format(self.__class__.__name__))
# -------------------------------------------------------------------------
# Internal-usage
# -------------------------------------------------------------------------
@property
def _are_restrained_atoms_defined(self):
"""Check if the restrained atoms are defined well enough to make a restraint"""
for atoms in [self.restrained_receptor_atoms, self.restrained_ligand_atoms]:
# Atoms should be a list or None at this point due to the _RestrainedAtomsProperty class
if atoms is None or not (isinstance(atoms, list) and len(atoms) > 0):
return False
return True
def _determine_restrained_atoms(self, sampler_state, topography):
"""Determine the atoms to restrain.
If the user has explicitly specified which atoms to restrained, this
does nothing, otherwise it picks the centroid of the receptor and
the centroid of the ligand as the two atoms to restrain.
Parameters
----------
sampler_state : openmmtools.states.SamplerState, optional
The sampler state holding the positions of all atoms.
topography : yank.Topography, optional
The topography with labeled receptor, ligand atoms, and any regions defined.
"""
debug_msg = ('Restraint {}: Automatically picked restrained '
'{{0}} atom: {{0}}'.format(self.__class__.__name__))
# No need to determine parameters if atoms have been given.
if self._are_restrained_atoms_defined:
return
# Shortcuts
positions = sampler_state.positions
# If receptor and ligand atoms are explicitly provided, use those.
restrained_ligand_atoms = self.restrained_ligand_atoms
restrained_receptor_atoms = self.restrained_receptor_atoms
@functools.singledispatch
def compute_atom_set(input_atoms, topography_key, mapping_function):
"""
Helper function for doing set operations on generic atom types.
mapping_function not used in the generic catch-all, but is used in the None register
"""
# Ensure the input atoms are only part of the topography_key atoms. Make no changes if they are
input_atoms_set = set(input_atoms)
set_topography_atoms = set(getattr(topography, topography_key))
intersect_set = input_atoms_set & set_topography_atoms
if intersect_set != input_atoms_set:
logger.warning("Some atoms specified by {0} were not actual {0}! "
"Atoms not part of {0} will be ignored.".format(topography_key))
final_atoms = list(intersect_set)
else:
final_atoms = list(input_atoms)
return final_atoms
@compute_atom_set.register(type(None))
def compute_atom_none(_, topography_key, mapping_function):
"""Helper for None type parsing"""
# Can't use list() here since mapping function returns a single integer.
atom_selection = [mapping_function(positions, getattr(topography, topography_key))]
logger.debug(debug_msg.format(topography_key, atom_selection))
return atom_selection
@compute_atom_set.register(str)
def compute_atom_str(input_string, topography_key, _):
"""Helper for string parsing"""
selection = topography.select(input_string, as_set=True)
selection_with_top = selection & set(getattr(topography, topography_key))
# Force output to be a normal int, dont need to worry about floats at this point, there should not be any
# If they come out as np.int64's, OpenMM complains
return [*map(int, selection_with_top)]
self.restrained_ligand_atoms = compute_atom_set(restrained_ligand_atoms,
'ligand_atoms',
self._closest_atom_to_centroid)
self.restrained_receptor_atoms = compute_atom_set(restrained_receptor_atoms,
'receptor_atoms',
self._closest_atom_to_centroid)
def _create_restraint_force(self, particles1, particles2):
"""Create a new restraint force between specified atoms.
Parameters
----------
particles1 : list of int
Indices of first group of atoms to restraint.
particles2 : list of int
Indices of second group of atoms to restraint.
Returns
-------
force : simtk.openmm.CustomBondForce
The created restraint force.
"""
# Check if parameters have been defined.
if self._bond_parameters is None:
err_msg = 'Restraint {}: Undefined bond parameters.'.format(self.__class__.__name__)
raise RestraintParameterError(err_msg)
# Unzip bond parameters names and values from dict.
parameter_names, parameter_values = zip(*self._bond_parameters.items())
# Create bond force and lambda_restraints parameter to control it.
if len(particles1) == 1 and len(particles2) == 1:
# CustomCentroidBondForce works only on 64bit platforms. When the
# restrained groups only have 1 particle, we can use the standard
# CustomBondForce so that we can support 32bit platforms too.
energy_function = self._energy_function.replace('distance(g1,g2)', 'r')
force = openmm.CustomBondForce('lambda_restraints * ' + energy_function)
force.addBond(particles1[0], particles2[0], parameter_values)
else:
force = openmm.CustomCentroidBondForce(2, 'lambda_restraints * ' + self._energy_function)
force.addGroup(particles1)
force.addGroup(particles2)
force.addBond([0, 1], parameter_values)
# Add all parameters.
force.addGlobalParameter('lambda_restraints', 1.0)
for parameter in parameter_names:
force.addPerBondParameter(parameter)
return force
@staticmethod
def _closest_atom_to_centroid(positions, indices=None, masses=None):
"""
Identify the closest atom to the centroid of the given coordinate set.
Parameters
----------
positions : unit.Quantity of natoms x 3 with units compatible with nanometers
positions of object to identify atom closes to centroid
indices : list of int, optional, default=None
List of atoms indices for which closest atom to centroid is to be computed.
masses : simtk.unit.Quantity of natoms with units compatible with amu
Masses of particles used to weight distance calculation, if not None (default: None)
Returns
-------
closest_atom : int
Index of atom closest to centroid of specified atoms.
"""
if indices is not None:
positions = positions[indices, :]
# Get dimensionless positions.
x_unit = positions.unit
x = positions / x_unit
# Determine number of atoms.
natoms = x.shape[0]
# Compute (natoms,1) array of normalized weights.
w = np.ones([natoms, 1])
if masses is not None:
w = masses / masses.unit # (natoms,) array
w = np.reshape(w, (natoms, 1)) # (natoms,1) array
w /= w.sum()
# Compute centroid (still in dimensionless units).
centroid = (np.tile(w, (1, 3)) * x).sum(0) # (3,) array
# Compute distances from centroid.
distances = np.sqrt(((x - np.tile(centroid, (natoms, 1)))**2).sum(1)) # distances[i] is the distance from the centroid to particle i
# Determine closest atom.
closest_atom = int(np.argmin(distances))
if indices is not None:
closest_atom = indices[closest_atom]
return closest_atom
# ==============================================================================
# Harmonic protein-ligand restraint.
# ==============================================================================
[docs]class Harmonic(RadiallySymmetricRestraint):
"""Impose a single harmonic restraint between ligand and protein.
This can be used to prevent the ligand from drifting too far from the
protein in implicit solvent calculations or to keep the ligand close
to the binding pocket in the decoupled states to increase mixing.
The restraint is applied between the centroids of two groups of atoms
that belong to the receptor and the ligand respectively. The centroids
are determined by a mass-weighted average of the group particles positions.
The energy expression of the restraint is given by
``E = lambda_restraints * (K/2)*r^2``
where `K` is the spring constant, `r` is the distance between the
two group centroids, and `lambda_restraints` is a scale factor that
can be used to control the strength of the restraint. You can control
``lambda_restraints`` through :class:`RestraintState` class.
The class supports automatic determination of the parameters left undefined or defined by strings
in the constructor through :func:`determine_missing_parameters`.
With OpenCL, groups with more than 1 atom are supported only on 64bit
platforms.
Parameters
----------
spring_constant : simtk.unit.Quantity, optional
The spring constant K (see energy expression above) in units compatible
with joule/nanometer**2/mole (default is None).
restrained_receptor_atoms : iterable of int, int, or str, optional
The indices of the receptor atoms to restrain, an MDTraj DSL expression, or a
:class:`Topography <yank.Topography>` region name,
or :func:`Topography Select String <yank.Topography.select>`.
This can temporarily be left undefined, but ``determine_missing_parameters()``
must be called before using the Restraint object. The same if a DSL
expression or Topography region is provided (default is None).
restrained_ligand_atoms : iterable of int, int, or str, optional
The indices of the ligand atoms to restrain, an MDTraj DSL expression.
or a :class:`Topography <yank.Topography>` region name,
or :func:`Topography Select String <yank.Topography.select>`.
This can temporarily be left undefined, but ``determine_missing_parameters()``
must be called before using the Restraint object. The same if a DSL
expression or Topography region is provided (default is None).
Attributes
----------
restrained_receptor_atoms : list of int, str, or None
The indices of the receptor atoms to restrain, an MDTraj selection string, or a Topography region selection
string.
restrained_ligand_atoms : list of int, str, or None
The indices of the ligand atoms to restrain, an MDTraj selection string, or a Topography region selection
string.
Examples
--------
Create the ThermodynamicState.
>>> from openmmtools import testsystems, states
>>> system_container = testsystems.LysozymeImplicit()
>>> system, positions = system_container.system, system_container.positions
>>> thermodynamic_state = states.ThermodynamicState(system, 300*unit.kelvin)
>>> sampler_state = states.SamplerState(positions)
Identify ligand atoms. Topography automatically identify receptor atoms too.
>>> from yank.yank import Topography
>>> topography = Topography(system_container.topology, ligand_atoms=range(2603, 2621))
you can create a completely defined restraint
>>> restraint = Harmonic(spring_constant=8*unit.kilojoule_per_mole/unit.nanometers**2,
... restrained_receptor_atoms=[1644, 1650, 1678],
... restrained_ligand_atoms='resname TMP')
Or automatically identify the parameters. When trying to impose a restraint
with undefined parameters, RestraintParameterError is raised.
>>> restraint = Harmonic()
>>> try:
... restraint.restrain_state(thermodynamic_state)
... except RestraintParameterError:
... print('There are undefined parameters. Choosing restraint parameters automatically.')
... restraint.determine_missing_parameters(thermodynamic_state, sampler_state, topography)
... restraint.restrain_state(thermodynamic_state)
...
There are undefined parameters. Choosing restraint parameters automatically.
Get standard state correction.
>>> correction = restraint.get_standard_state_correction(thermodynamic_state)
"""
def __init__(self, spring_constant=None, **kwargs):
super(Harmonic, self).__init__(**kwargs)
self.spring_constant = spring_constant
@property
def _energy_function(self):
"""str: energy expression of the restraint force."""
return '(K/2)*distance(g1,g2)^2'
@property
def _bond_parameters(self):
"""dict: the bond parameters of the restraint force.
If there are parameters undefined, this is None.
"""
if self.spring_constant is None:
return None
return {'K': self.spring_constant}
def _determine_bond_parameters(self, thermodynamic_state, sampler_state, topography):
"""Automatically choose a spring constant for the restraint force.
The spring constant is selected to give 1 kT at one standard deviation
of receptor atoms about the receptor restrained atom.
Parameters
----------
thermodynamic_state : openmmtools.states.ThermodynamicState
The thermodynamic state.
sampler_state : openmmtools.states.SamplerState
The sampler state holding the positions of all atoms.
topography : yank.Topography
The topography with labeled receptor and ligand atoms.
"""
# Do not overwrite parameters that are already defined.
if self.spring_constant is not None:
return
receptor_positions = sampler_state.positions[topography.receptor_atoms]
sigma = pipeline.compute_radius_of_gyration(receptor_positions)
# Compute corresponding spring constant.
self.spring_constant = thermodynamic_state.kT / sigma**2
logger.debug('Spring constant sigma, s = {:.3f} nm'.format(sigma / unit.nanometers))
logger.debug('K = {:.1f} kcal/mol/A^2'.format(
self.spring_constant / unit.kilocalories_per_mole * unit.angstroms**2))
# ==============================================================================
# Flat-bottom protein-ligand restraint.
# ==============================================================================
[docs]class FlatBottom(RadiallySymmetricRestraint):
"""A receptor-ligand restraint using a flat potential well with harmonic walls.
An alternative choice to receptor-ligand restraints that uses a flat
potential inside most of the protein volume with harmonic restraining
walls outside of this. It can be used to prevent the ligand from
drifting too far from protein in implicit solvent calculations while
still exploring the surface of the protein for putative binding sites.
The restraint is applied between the centroids of two groups of atoms
that belong to the receptor and the ligand respectively. The centroids
are determined by a mass-weighted average of the group particles positions.
More precisely, the energy expression of the restraint is given by
``E = lambda_restraints * step(r-r0) * (K/2)*(r-r0)^2``
where ``K`` is the spring constant, ``r`` is the distance between the
restrained atoms, ``r0`` is another parameter defining the distance
at which the restraint is imposed, and ``lambda_restraints``
is a scale factor that can be used to control the strength of the
restraint. You can control ``lambda_restraints`` through the class
:class:`RestraintState`.
The class supports automatic determination of the parameters left undefined
in the constructor through :func:`determine_missing_parameters`.
With OpenCL, groups with more than 1 atom are supported only on 64bit
platforms.
Parameters
----------
spring_constant : simtk.unit.Quantity, optional
The spring constant K (see energy expression above) in units compatible
with joule/nanometer**2/mole (default is None).
well_radius : simtk.unit.Quantity, optional
The distance r0 (see energy expression above) at which the harmonic
restraint is imposed in units of distance (default is None).
restrained_receptor_atoms : iterable of int, int, or str, optional
The indices of the receptor atoms to restrain, an MDTraj DSL expression, or a
:class:`Topography <yank.Topography>` region name,
or :func:`Topography Select String <yank.Topography.select>`.
This can temporarily be left undefined, but ``determine_missing_parameters()``
must be called before using the Restraint object. The same if a DSL
expression or Topography region is provided (default is None).
restrained_ligand_atoms : iterable of int, int, or str, optional
The indices of the ligand atoms to restrain, an MDTraj DSL expression.
or a :class:`Topography <yank.Topography>` region name,
or :func:`Topography Select String <yank.Topography.select>`.
This can temporarily be left undefined, but ``determine_missing_parameters()``
must be called before using the Restraint object. The same if a DSL
expression or Topography region is provided (default is None).
Attributes
----------
restrained_receptor_atoms : list of int or None
The indices of the receptor atoms to restrain, an MDTraj selection string, or a Topography region selection
string.
restrained_ligand_atoms : list of int or None
The indices of the ligand atoms to restrain, an MDTraj selection string, or a Topography region selection
string.
Examples
--------
Create the ThermodynamicState.
>>> from openmmtools import testsystems, states
>>> system_container = testsystems.LysozymeImplicit()
>>> system, positions = system_container.system, system_container.positions
>>> thermodynamic_state = states.ThermodynamicState(system, 298*unit.kelvin)
>>> sampler_state = states.SamplerState(positions)
Identify ligand atoms. Topography automatically identify receptor atoms too.
>>> from yank.yank import Topography
>>> topography = Topography(system_container.topology, ligand_atoms=range(2603, 2621))
You can create a completely defined restraint
>>> restraint = FlatBottom(spring_constant=0.6*unit.kilocalorie_per_mole/unit.angstroms**2,
... well_radius=5.2*unit.nanometers, restrained_receptor_atoms=[1644, 1650, 1678],
... restrained_ligand_atoms='resname TMP')
or automatically identify the parameters. When trying to impose a restraint
with undefined parameters, RestraintParameterError is raised.
>>> restraint = FlatBottom()
>>> try:
... restraint.restrain_state(thermodynamic_state)
... except RestraintParameterError:
... print('There are undefined parameters. Choosing restraint parameters automatically.')
... restraint.determine_missing_parameters(thermodynamic_state, sampler_state, topography)
... restraint.restrain_state(thermodynamic_state)
...
There are undefined parameters. Choosing restraint parameters automatically.
Get standard state correction.
>>> correction = restraint.get_standard_state_correction(thermodynamic_state)
"""
def __init__(self, spring_constant=None, well_radius=None, **kwargs):
super(FlatBottom, self).__init__(**kwargs)
self.spring_constant = spring_constant
self.well_radius = well_radius
@property
def _energy_function(self):
"""str: energy expression of the restraint force."""
return 'step(distance(g1,g2)-r0) * (K/2)*(distance(g1,g2)-r0)^2'
@property
def _bond_parameters(self):
"""dict: the bond parameters of the restraint force.
If there are parameters undefined, this is None.
"""
if self.spring_constant is None or self.well_radius is None:
return None
return {'K': self.spring_constant, 'r0': self.well_radius}
def _determine_bond_parameters(self, thermodynamic_state, sampler_state, topography):
"""Automatically choose a spring constant and well radius.
The spring constant, is set to 5.92 kcal/mol/A**2, the well
radius is set at twice the robust estimate of the standard
deviation (from mean absolute deviation) plus 5 A.
Parameters
----------
thermodynamic_state : openmmtools.states.ThermodynamicState
The thermodynamic state.
sampler_state : openmmtools.states.SamplerState
The sampler state holding the positions of all atoms.
topography : yank.Topography
The topography with labeled receptor and ligand atoms.
"""
# Determine number of atoms.
n_atoms = len(topography.receptor_atoms)
# Check that restrained receptor atoms are in expected range.
if any(atom_id >= n_atoms for atom_id in self.restrained_receptor_atoms):
raise ValueError('Receptor atoms {} were selected for restraint, but system '
'only has {} atoms.'.format(self.restrained_receptor_atoms, n_atoms))
# Compute well radius if the user hasn't specified it in the constructor.
if self.well_radius is None:
# Get positions of mass-weighted centroid atom.
# (Working in non-unit-bearing floats for speed.)
x_unit = sampler_state.positions.unit
x_restrained_atoms = sampler_state.positions[self.restrained_receptor_atoms, :] / x_unit
system = thermodynamic_state.system
masses = np.array([system.getParticleMass(i) / unit.dalton for i in self.restrained_receptor_atoms])
x_centroid = np.average(x_restrained_atoms, axis=0, weights=masses)
# Get dimensionless receptor and ligand positions.
x_receptor = sampler_state.positions[topography.receptor_atoms, :] / x_unit
x_ligand = sampler_state.positions[topography.ligand_atoms, :] / x_unit
# Compute maximum square distance from the centroid to any receptor atom.
# dist2_centroid_receptor[i] is the squared distance from the centroid to receptor atom i.
dist2_centroid_receptor = pipeline.compute_squared_distances([x_centroid], x_receptor)
max_dist_receptor = np.sqrt(dist2_centroid_receptor.max()) * x_unit
# Compute maximum length of the ligand. dist2_ligand_ligand[i][j] is the
# squared distance between atoms i and j of the ligand.
dist2_ligand_ligand = pipeline.compute_squared_distances(x_ligand, x_ligand)
max_length_ligand = np.sqrt(dist2_ligand_ligand.max()) * x_unit
# Compute the radius of the flat bottom restraint.
self.well_radius = max_dist_receptor + max_length_ligand/2 + 5*unit.angstrom
# Set default spring constant if the user hasn't specified it in the constructor.
if self.spring_constant is None:
self.spring_constant = 10.0 * thermodynamic_state.kT / unit.angstroms**2
logger.debug('restraint distance r0 = {:.1f} A'.format(self.well_radius / unit.angstroms))
logger.debug('K = {:.1f} kcal/mol/A^2'.format(
self.spring_constant / unit.kilocalories_per_mole * unit.angstroms**2))
# ==============================================================================
# Orientation-dependent receptor-ligand restraints.
# ==============================================================================
[docs]class Boresch(ReceptorLigandRestraint):
"""Impose Boresch-style orientational restraints on protein-ligand system.
This restraints the ligand binding mode by constraining 1 distance, 2
angles and 3 dihedrals between 3 atoms of the receptor and 3 atoms of
the ligand.
More precisely, the energy expression of the restraint is given by
.. code-block:: python
E = lambda_restraints * {
K_r/2 * [|r3 - l1| - r_aA0]^2 +
+ K_thetaA/2 * [angle(r2,r3,l1) - theta_A0]^2 +
+ K_thetaB/2 * [angle(r3,l1,l2) - theta_B0]^2 +
+ K_phiA/2 * [dihedral(r1,r2,r3,l1) - phi_A0]^2 +
+ K_phiB/2 * [dihedral(r2,r3,l1,l2) - phi_B0]^2 +
+ K_phiC/2 * [dihedral(r3,l1,l2,l3) - phi_C0]^2
}
, where the parameters are:
``r1``, ``r2``, ``r3``: the coordinates of the 3 receptor atoms.
``l1``, ``l2``, ``l3``: the coordinates of the 3 ligand atoms.
``K_r``: the spring constant for the restrained distance ``|r3 - l1|``.
``r_aA0``: the equilibrium distance of ``|r3 - l1|``.
``K_thetaA``, ``K_thetaB``: the spring constants for ``angle(r2,r3,l1)`` and ``angle(r3,l1,l2)``.
``theta_A0``, ``theta_B0``: the equilibrium angles of ``angle(r2,r3,l1)`` and ``angle(r3,l1,l2)``.
``K_phiA``, ``K_phiB``, ``K_phiC``: the spring constants for ``dihedral(r1,r2,r3,l1)``,
``dihedral(r2,r3,l1,l2)``, ``dihedral(r3,l1,l2,l3)``.
``phi_A0``, ``phi_B0``, ``phi_C0``: the equilibrium torsion of ``dihedral(r1,r2,r3,l1)``,
``dihedral(r2,r3,l1,l2)``, ``dihedral(r3,l1,l2,l3)``.
``lambda_restraints``: a scale factor that can be used to control the strength
of the restraint.
You can control ``lambda_restraints`` through the class :class:`RestraintState`.
The class supports automatic determination of the parameters left undefined
in the constructor through :func:`determine_missing_parameters`.
*Warning*: Symmetry corrections for symmetric ligands are not automatically applied.
See Ref [1] and [2] for more information on correcting for ligand symmetry.
*Warning*: Only heavy atoms can be restrained. Hydrogens will automatically be excluded.
Parameters
----------
restrained_receptor_atoms : iterable of int, str, or None; Optional
The indices of the receptor atoms to restrain, an MDTraj DSL expression, or a
:class:`Topography <yank.Topography>` region name,
or :func:`Topography Select String <yank.Topography.select>`.
If this is a list of three ints, the receptor atoms will be restrained in order, r1, r2, r3. If there are more
than three entries or the selection string resolves more than three atoms, the three restrained atoms will
be chosen at random from the selection.
This can temporarily be left undefined, but ``determine_missing_parameters()``
must be called before using the Restraint object. The same if a DSL
expression or Topography region is provided (default is None).
restrained_ligand_atoms : iterable of int, str, or None; Optional
The indices of the ligand atoms to restrain, an MDTraj DSL expression, or a
:class:`Topography <yank.Topography>` region name,
or :func:`Topography Select String <yank.Topography.select>`.
If this is a list of three ints, the receptor atoms will be restrained in order, l1, l2, l3. If there are more
than three entries or the selection string resolves more than three atoms, the three restrained atoms will
be chosen at random from the selection.
This can temporarily be left undefined, but ``determine_missing_parameters()``
must be called before using the Restraint object. The same if a DSL
expression or Topography region is provided (default is None).
K_r : simtk.unit.Quantity, optional
The spring constant for the restrained distance ``|r3 - l1|`` (units
compatible with kilocalories_per_mole/angstrom**2).
r_aA0 : simtk.unit.Quantity, optional
The equilibrium distance between r3 and l1 (units of length).
K_thetaA, K_thetaB : simtk.unit.Quantity, optional
The spring constants for ``angle(r2, r3, l1)`` and ``angle(r3, l1, l2)``
(units compatible with kilocalories_per_mole/radians**2).
theta_A0, theta_B0 : simtk.unit.Quantity, optional
The equilibrium angles of ``angle(r2, r3, l1)`` and ``angle(r3, l1, l2)``
(units compatible with radians).
K_phiA, K_phiB, K_phiC : simtk.unit.Quantity, optional
The spring constants for ``dihedral(r1, r2, r3, l1)``,
``dihedral(r2, r3, l1, l2)`` and ``dihedral(r3,l1,l2,l3)`` (units compatible
with kilocalories_per_mole/radians**2).
phi_A0, phi_B0, phi_C0 : simtk.unit.Quantity, optional
The equilibrium torsion of ``dihedral(r1,r2,r3,l1)``, ``dihedral(r2,r3,l1,l2)``
and ``dihedral(r3,l1,l2,l3)`` (units compatible with radians).
standard_state_correction_method : 'analytical' or 'numeric', optional
The method to use to estimate the standard state correction (default
is 'analytical').
Attributes
----------
restrained_receptor_atoms : list of int
The indices of the 3 receptor atoms to restrain [r1, r2, r3].
restrained_ligand_atoms : list of int
The indices of the 3 ligand atoms to restrain [l1, l2, l3].
standard_state_correction_method
References
----------
[1] Boresch S, Tettinger F, Leitgeb M, Karplus M. J Phys Chem B. 107:9535, 2003.
http://dx.doi.org/10.1021/jp0217839
[2] Mobley DL, Chodera JD, and Dill KA. J Chem Phys 125:084902, 2006.
https://dx.doi.org/10.1063%2F1.2221683
Examples
--------
Create the ThermodynamicState.
>>> from openmmtools import testsystems, states
>>> system_container = testsystems.LysozymeImplicit()
>>> system, positions = system_container.system, system_container.positions
>>> thermodynamic_state = states.ThermodynamicState(system, 298*unit.kelvin)
>>> sampler_state = states.SamplerState(positions)
Identify ligand atoms. Topography automatically identify receptor atoms too.
>>> from yank.yank import Topography
>>> topography = Topography(system_container.topology, ligand_atoms=range(2603, 2621))
Create a partially defined restraint
>>> restraint = Boresch(restrained_receptor_atoms=[1335, 1339, 1397],
... restrained_ligand_atoms=[2609, 2607, 2606],
... K_r=20.0*unit.kilocalories_per_mole/unit.angstrom**2,
... r_aA0=0.35*unit.nanometer)
and automatically identify the other parameters. When trying to impose
a restraint with undefined parameters, RestraintParameterError is raised.
>>> try:
... restraint.restrain_state(thermodynamic_state)
... except RestraintParameterError:
... print('There are undefined parameters. Choosing restraint parameters automatically.')
... restraint.determine_missing_parameters(thermodynamic_state, sampler_state, topography)
... restraint.restrain_state(thermodynamic_state)
...
There are undefined parameters. Choosing restraint parameters automatically.
Get standard state correction.
>>> correction = restraint.get_standard_state_correction(thermodynamic_state)
"""
def __init__(self, restrained_receptor_atoms=None, restrained_ligand_atoms=None,
K_r=None, r_aA0=None,
K_thetaA=None, theta_A0=None,
K_thetaB=None, theta_B0=None,
K_phiA=None, phi_A0=None,
K_phiB=None, phi_B0=None,
K_phiC=None, phi_C0=None,
standard_state_correction_method='analytical'):
self.restrained_receptor_atoms = restrained_receptor_atoms
self.restrained_ligand_atoms = restrained_ligand_atoms
self.K_r = K_r
self.r_aA0 = r_aA0
self.K_thetaA, self.K_thetaB = K_thetaA, K_thetaB
self.theta_A0, self.theta_B0 = theta_A0, theta_B0
self.K_phiA, self.K_phiB, self.K_phiC = K_phiA, K_phiB, K_phiC
self.phi_A0, self.phi_B0, self.phi_C0 = phi_A0, phi_B0, phi_C0
self.standard_state_correction_method = standard_state_correction_method
# -------------------------------------------------------------------------
# Public properties.
# -------------------------------------------------------------------------
class _BoreschRestrainedAtomsProperty(_RestrainedAtomsProperty):
"""
Descriptor of restrained atoms.
Extends `_RestrainedAtomsProperty` to handle single integers and strings.
"""
_MUST_COMPUTE_STRING = ('You are specifying {} {} atoms, '
'the final atoms will be chosen at from this set but you MUST '
'run "determine_missing_parameters"')
@methoddispatch
def _validate_atoms(self, restrained_atoms):
restrained_atoms = super()._validate_atoms(restrained_atoms)
if len(restrained_atoms) < 3:
raise ValueError('At least three {} atoms are required to impose a '
'Boresch-style restraint.'.format(self._atoms_type))
elif len(restrained_atoms) > 3:
logger.warning(self._MUST_COMPUTE_STRING.format("more than three", self._atoms_type))
return restrained_atoms
@_validate_atoms.register(str)
def _cast_atom_string(self, restrained_atoms):
logger.warning(self._MUST_COMPUTE_STRING.format("a string for", self._atoms_type))
return restrained_atoms
restrained_receptor_atoms = _BoreschRestrainedAtomsProperty('receptor')
restrained_ligand_atoms = _BoreschRestrainedAtomsProperty('ligand')
@property
def standard_state_correction_method(self):
"""str: The default method to use in :func:`get_standard_state_correction`.
This can be either 'analytical' or 'numerical'.
"""
return self._standard_state_correction_method
@standard_state_correction_method.setter
def standard_state_correction_method(self, value):
if value not in ['analytical', 'numerical']:
raise ValueError("The standard state correction method must be one between "
"'analytical' and 'numerical', got {}.".format(value))
self._standard_state_correction_method = value
# -------------------------------------------------------------------------
# Public methods.
# -------------------------------------------------------------------------
[docs] def restrain_state(self, thermodynamic_state):
"""Add the restraint force to the state's ``System``.
Parameters
----------
thermodynamic_state : openmmtools.states.ThermodynamicState
The thermodynamic state holding the system to modify.
"""
# TODO replace dihedral restraints with negative log von Mises distribution?
# https://en.wikipedia.org/wiki/Von_Mises_distribution, the von Mises parameter
# kappa would be computed from the desired standard deviation (kappa ~ sigma**(-2))
# and the standard state correction would need to be modified.
# Check if all parameters are defined.
self._check_parameters_defined()
energy_function = """
lambda_restraints * E;
E = (K_r/2)*(distance(p3,p4) - r_aA0)^2
+ (K_thetaA/2)*(angle(p2,p3,p4)-theta_A0)^2 + (K_thetaB/2)*(angle(p3,p4,p5)-theta_B0)^2
+ (K_phiA/2)*dphi_A^2 + (K_phiB/2)*dphi_B^2 + (K_phiC/2)*dphi_C^2;
dphi_A = dA - floor(dA/(2*pi)+0.5)*(2*pi); dA = dihedral(p1,p2,p3,p4) - phi_A0;
dphi_B = dB - floor(dB/(2*pi)+0.5)*(2*pi); dB = dihedral(p2,p3,p4,p5) - phi_B0;
dphi_C = dC - floor(dC/(2*pi)+0.5)*(2*pi); dC = dihedral(p3,p4,p5,p6) - phi_C0;
pi = %f;
""" % np.pi
# Add constant definitions to the energy function
for name, value in self._parameters.items():
energy_function += '%s = %f; ' % (name, value.value_in_unit_system(unit.md_unit_system))
# Create the force
n_particles = 6 # number of particles involved in restraint: p1 ... p6
restraint_force = openmm.CustomCompoundBondForce(n_particles, energy_function)
restraint_force.addGlobalParameter('lambda_restraints', 1.0)
restraint_force.addBond(self.restrained_receptor_atoms + self.restrained_ligand_atoms, [])
restraint_force.setUsesPeriodicBoundaryConditions(thermodynamic_state.is_periodic)
# Get a copy of the system of the ThermodynamicState, modify it and set it back.
system = thermodynamic_state.system
system.addForce(restraint_force)
thermodynamic_state.system = system
[docs] def get_standard_state_correction(self, thermodynamic_state):
"""Return the standard state correction.
Parameters
----------
thermodynamic_state : openmmtools.states.ThermodynamicState
The thermodynamic state.
Returns
-------
DeltaG : float
Computed standard-state correction in dimensionless units (kT).
"""
if self.standard_state_correction_method == 'analytical':
return self._get_standard_state_correction_analytical(thermodynamic_state)
else: # The property checks that the value is known in the setter.
return self._get_standard_state_correction_numerical(thermodynamic_state)
[docs] def determine_missing_parameters(self, thermodynamic_state, sampler_state, topography):
"""Determine parameters and restrained atoms automatically.
Currently, all equilibrium values are measured from the initial structure,
while spring constants are set to 20 kcal/(mol A**2) or 20 kcal/(mol rad**2)
as in Ref [1]. The restrained atoms are selected so that the analytical
standard state correction will be valid.
Parameters that have been already specified are left untouched.
Future iterations of this feature will introduce the ability to extract
equilibrium parameters and spring constants from a short simulation.
Parameters
----------
thermodynamic_state : openmmtools.states.ThermodynamicState
The thermodynamic state.
sampler_state : openmmtools.states.SamplerState, optional
The sampler state holding the positions of all atoms.
topography : yank.Topography, optional
The topography with labeled receptor and ligand atoms.
"""
MAX_ATTEMPTS = 100
logger.debug('Automatically selecting restraint atoms and parameters:')
# If restrained atoms are already specified, we only need to determine parameters.
if self._are_restrained_atoms_defined:
self._determine_restraint_parameters(sampler_state, topography)
else:
# Keep selecting random retrained atoms until the parameters
# make the standard state correction robust.
for attempt in range(MAX_ATTEMPTS):
logger.debug('Attempt {} / {} at automatically selecting atoms and '
'restraint parameters...'.format(attempt, MAX_ATTEMPTS))
# Randomly pick non-collinear atoms.
restrained_atoms = self._pick_restrained_atoms(sampler_state, topography)
self.restrained_receptor_atoms = restrained_atoms[:3]
self.restrained_ligand_atoms = restrained_atoms[3:]
# Determine restraint parameters for these atoms.
self._determine_restraint_parameters(sampler_state, topography)
# Check if we have found a good solution.
if self._is_analytical_correction_robust(thermodynamic_state.kT) is True:
break
# Check if the analytical standard state correction is robust with these parameters.
# This check is must be performed both in the case where the user has provided the
# restrained atoms, and in the case where we exhausted the number of attempts.
if (not self._is_analytical_correction_robust(thermodynamic_state.kT) and
self.standard_state_correction_method == 'analytical'):
logger.warning('The provided restrained atoms do not guarantee a robust calculation of '
'the standard state correction. Switching to the numerical scheme.')
self.standard_state_correction_method = 'numerical'
# -------------------------------------------------------------------------
# Internal-usage
# -------------------------------------------------------------------------
@property
def _parameters(self):
"""dict: restraint parameters in dict forms."""
parameter_names, _, _, _ = inspect.getargspec(self.__init__)
# Exclude non-parameters arguments.
for exclusion in ['self', 'restrained_receptor_atoms', 'restrained_ligand_atoms',
'standard_state_correction_method']:
parameter_names.remove(exclusion)
# Retrieve and store options.
parameters = {parameter_name: getattr(self, parameter_name)
for parameter_name in parameter_names}
return parameters
def _check_parameters_defined(self):
"""Raise an exception there are still parameters undefined."""
if not self._are_restrained_atoms_defined:
raise RestraintParameterError('Undefined restrained atoms.')
# Find undefined parameters and raise error.
undefined_parameters = [name for name, value in self._parameters.items() if value is None]
if len(undefined_parameters) > 0:
err_msg = 'Undefined parameters for Boresch restraint: {}'.format(undefined_parameters)
raise RestraintParameterError(err_msg)
@property
def _are_restrained_atoms_defined(self):
"""Check if the restrained atoms are defined well enough to make a restraint"""
for atoms in [self.restrained_receptor_atoms, self.restrained_ligand_atoms]:
# Atoms should be a list or None at this point due to the _RestrainedAtomsProperty class
if atoms is None or not (isinstance(atoms, list) and len(atoms) == 3):
return False
return True
def _is_analytical_correction_robust(self, kT):
"""Check that the analytical standard state correction is valid for the current parameters."""
N_SIGMA = 4
is_robust = True
for name in ['A', 'B']:
theta0 = getattr(self, 'theta_' + name + '0')
K = getattr(self, 'K_theta' + name)
sigma = unit.sqrt(N_SIGMA * kT * 2.0 / K)
if theta0 < sigma or theta0 > np.pi*unit.radians - sigma:
logger.debug('theta_' + name + '0 is too close to 0 or pi '
'for standard state correction to be accurate.')
is_robust = False
r0 = getattr(self, 'r_aA0')
K = getattr(self, 'K_r')
sigma = unit.sqrt(N_SIGMA * kT * 2.0 / K)
if r0 < sigma:
logger.debug('r_aA0 is too close to 0 for standard state correction to be accurate.')
is_robust = False
return is_robust
def _get_standard_state_correction_analytical(self, thermodynamic_state):
"""Return the standard state correction using the analytical method.
Uses analytical approach from [1], but this approach is known to be inexact.
This approach breaks down when the equilibrium restraint angles are near the
limits of their domains and when equilibrium distance is near 0.
Parameters
----------
thermodynamic_state : openmmtools.states.ThermodynamicState
The thermodynamic state.
Returns
-------
DeltaG : float
Computed standard-state correction in dimensionless units (kT).
"""
# Check if all parameters are defined.
self._check_parameters_defined()
# Shortcuts variables.
pi = np.pi
kT = thermodynamic_state.kT
p = self # For the parameters.
# Eq 32 of Ref [1]. Multiply by unit.radian**5 to remove the
# expected unit value radians is a soft unit in this case, it
# cancels in the math, but not in the equations above.
DeltaG = -np.log(
(8. * pi ** 2 * V0) / (p.r_aA0 ** 2 * unit.sin(p.theta_A0) * unit.sin(p.theta_B0)) *
unit.sqrt(p.K_r * p.K_thetaA * p.K_thetaB * p.K_phiA * p.K_phiB * p.K_phiC) / (2 * pi * kT) ** 3 *
unit.radian**5
)
return DeltaG
def _get_standard_state_correction_numerical(self, thermodynamic_state):
"""Return the standard state correction using the numerical method.
Uses numerical integral to the partition function contributions for
r and theta, analytical for phi
Parameters
----------
thermodynamic_state : openmmtools.states.ThermodynamicState
The thermodynamic state.
Returns
-------
DeltaG : float
Computed standard-state correction in dimensionless units (kT).
"""
# Check if all parameters are defined.
self._check_parameters_defined()
def strip(passed_unit):
"""Cast the passed_unit into md unit system for integrand lambda functions"""
return passed_unit.value_in_unit_system(unit.md_unit_system)
# Shortcuts variables.
pi = np.pi
kT = thermodynamic_state.kT
p = self # For the parameters.
# Radial
sigma = 1 / unit.sqrt(p.K_r / kT)
rmin = min(0*unit.angstrom, p.r_aA0 - 8 * sigma)
rmax = p.r_aA0 + 8 * sigma
I = lambda r: r ** 2 * np.exp(-strip(p.K_r) / (2 * strip(kT)) * (r - strip(p.r_aA0)) ** 2)
DGIntegral, dDGIntegral = scipy.integrate.quad(I, strip(rmin), strip(rmax)) * unit.nanometer**3
ExpDeltaG = DGIntegral
# Angular
for name in ['A', 'B']:
theta0 = getattr(p, 'theta_' + name + '0')
K_theta = getattr(p, 'K_theta' + name)
I = lambda theta: np.sin(theta) * np.exp(-strip(K_theta) / (2 * strip(kT)) * (theta - strip(theta0)) ** 2)
DGIntegral, dDGIntegral = scipy.integrate.quad(I, 0, pi)
ExpDeltaG *= DGIntegral
# Torsion
for name in ['A', 'B', 'C']:
phi0 = getattr(p, 'phi_' + name + '0')
K_phi = getattr(p, 'K_phi' + name)
kshort = strip(K_phi/kT)
ExpDeltaG *= math.sqrt(pi/2.0) * (
math.erf((strip(phi0)+pi)*unit.sqrt(kshort)/math.sqrt(2)) -
math.erf((strip(phi0)-pi)*unit.sqrt(kshort)/math.sqrt(2))
) / unit.sqrt(kshort)
DeltaG = -np.log(8 * pi**2 * V0 / ExpDeltaG)
return DeltaG
@staticmethod
def _is_collinear(positions, atoms, threshold=0.9):
"""Report whether any sequential vectors in a sequence of atoms are collinear.
Parameters
----------
positions : n_atoms x 3 simtk.unit.Quantity
Reference positions to use for imposing restraints (units of length).
atoms : iterable of int
The indices of the atoms to test.
threshold : float, optional, default=0.9
Atoms are not collinear if their sequential vector separation dot
products are less than ``threshold``.
Returns
-------
result : bool
Returns True if any sequential pair of vectors is collinear; False otherwise.
"""
result = False
for i in range(len(atoms)-2):
v1 = positions[atoms[i+1], :] - positions[atoms[i], :]
v2 = positions[atoms[i+2], :] - positions[atoms[i+1], :]
normalized_inner_product = np.dot(v1, v2) / np.sqrt(np.dot(v1, v1) * np.dot(v2, v2))
result = result or (normalized_inner_product > threshold)
return result
def _pick_restrained_atoms(self, sampler_state, topography):
"""Select atoms to be used in restraint.
Parameters
----------
sampler_state : openmmtools.states.SamplerState, optional
The sampler state holding the positions of all atoms.
topography : yank.Topography, optional
The topography with labeled receptor and ligand atoms.
Returns
-------
restrained_atoms : list of int
List of six atom indices used in the restraint.
restrained_atoms[0:3] belong to the receptor,
restrained_atoms[4:6] belong to the ligand.
Notes
-----
The current algorithm simply selects random subsets of receptor
and ligand atoms and rejects those that are too close to collinear.
Future updates can further refine this algorithm.
"""
# No need to determine parameters if atoms have been given.
if self._are_restrained_atoms_defined:
return self.restrained_receptor_atoms + self.restrained_ligand_atoms
# If receptor and ligand atoms are explicitly provided, use those.
heavy_ligand_atoms = self.restrained_ligand_atoms
heavy_receptor_atoms = self.restrained_receptor_atoms
# Otherwise we restrain only heavy atoms.
heavy_atoms = set(topography.topology.select('not element H').tolist())
# Intersect heavy atoms with receptor/ligand atoms (s1&s2 is intersect).
atom_inclusion_warning = ("Some atoms specified by {0} were not actual {0} and heavy atoms! "
"Atoms not meeting these criteria will be ignored.")
@functools.singledispatch
def compute_atom_set(input_atoms, topography_key):
"""Helper function for doing set operations on heavy ligand atoms of all other types"""
# If the length is 3, we don't want to make ANY changes, so don't modify the set
input_set = set(input_atoms)
topography_set = set(getattr(topography, topography_key))
intersect_set = input_set & heavy_atoms & topography_set
if intersect_set != input_set:
logger.warning(atom_inclusion_warning.format(topography_key))
return intersect_set
else:
# The return types are intentionally different types to handle some r3-l1 logic later
return input_atoms
@compute_atom_set.register(type(None))
def compute_atom_none(_, topography_key):
"""Helper for None type parsing"""
return set(getattr(topography, topography_key)) & heavy_atoms
@compute_atom_set.register(str)
def compute_atom_str(input_string, topography_key):
"""Helper for string parsing"""
output = topography.select(topography_key, as_set=False) # Preserve order
set_output = set(output)
set_topography = set(getattr(topography, topography_key))
# Ensure the selection is in the correct set
set_combined = set_output & set_topography & heavy_atoms
final_output = [particle for particle in output if particle in set_combined]
if len(final_output) < len(output):
logger.warning(atom_inclusion_warning.format(topography_key))
# Force output to be a normal int, dont need to worry about floats at this point, there should not be any
# If they come out as np.int64's, OpenMM complains
return [*map(int, final_output)]
heavy_ligand_atoms = compute_atom_set(heavy_ligand_atoms, 'ligand_atoms')
heavy_receptor_atoms = compute_atom_set(heavy_receptor_atoms, 'receptor_atoms')
if len(heavy_receptor_atoms) < 3 or len(heavy_ligand_atoms) < 3:
raise ValueError('There must be at least three heavy atoms in receptor_atoms '
'(# heavy {}) and ligand_atoms (# heavy {}).'.format(
len(heavy_receptor_atoms), len(heavy_ligand_atoms)))
# If r3 or l1 atoms are given. We have to pick those.
if isinstance(heavy_receptor_atoms, list):
r3_atoms = [heavy_receptor_atoms[2]]
else:
r3_atoms = heavy_receptor_atoms
if isinstance(heavy_ligand_atoms, list):
l1_atoms = [heavy_ligand_atoms[0]]
else:
l1_atoms = heavy_ligand_atoms
# TODO: Cast itertools generator to np array more efficiently
r3_l1_pairs = np.array(list(itertools.product(r3_atoms, l1_atoms)))
# Filter r3-l1 pairs that are too close/far away for the distance constraint.
max_distance = 4 * unit.angstrom/unit.nanometer
min_distance = 1 * unit.angstrom/unit.nanometer
t = md.Trajectory(sampler_state.positions / unit.nanometers, topography.topology)
distances = md.geometry.compute_distances(t, r3_l1_pairs)[0]
indices_of_in_range_pairs = np.where(np.logical_and(distances > min_distance, distances <= max_distance))[0]
if len(indices_of_in_range_pairs) == 0:
error_msg = ('There are no heavy ligand atoms within the range of [{},{}] nm heavy receptor atoms!\n'
'Please Check your input files or try another restraint class')
raise ValueError(error_msg.format(min_distance, max_distance))
r3_l1_pairs = r3_l1_pairs[indices_of_in_range_pairs].tolist()
# Iterate until we have found a set of non-collinear atoms.
accepted = False
while not accepted:
# Select a receptor/ligand atom in range of each other for the distance constraint.
r3_l1_atoms = random.sample(r3_l1_pairs, 1)[0]
r3_l1_atoms_set = set(r3_l1_atoms)
# Determine remaining receptor/ligand atoms.
if isinstance(heavy_receptor_atoms, list):
r1_r2_atoms = heavy_receptor_atoms[:2]
else:
r1_r2_atoms = random.sample(heavy_receptor_atoms - r3_l1_atoms_set, 2)
if isinstance(heavy_ligand_atoms, list):
l2_l3_atoms = heavy_ligand_atoms[1:]
else:
l2_l3_atoms = random.sample(heavy_ligand_atoms - r3_l1_atoms_set, 2)
# Reject collinear sets of atoms.
restrained_atoms = r1_r2_atoms + r3_l1_atoms + l2_l3_atoms
accepted = not self._is_collinear(sampler_state.positions, restrained_atoms)
logger.debug('Selected atoms to restrain: {}'.format(restrained_atoms))
return restrained_atoms
def _determine_restraint_parameters(self, sampler_states, topography):
"""Determine restraint parameters.
Currently, all equilibrium values are measured from the initial structure,
while spring constants are set to 20 kcal/(mol A**2) or 20 kcal/(mol rad**2)
as in [1].
Future iterations of this feature will introduce the ability to extract
equilibrium parameters and spring constants from a short simulation.
References
----------
[1] Boresch S, Tettinger F, Leitgeb M, Karplus M. J Phys Chem B. 107:9535, 2003.
http://dx.doi.org/10.1021/jp0217839
"""
# We determine automatically only the parameters that have been left undefined.
def _assign_if_undefined(attr_name, attr_value):
"""Assign value to self.name only if it is None."""
if getattr(self, attr_name) is None:
setattr(self, attr_name, attr_value)
# Merge receptor and ligand atoms in a single array for easy manipulation.
restrained_atoms = self.restrained_receptor_atoms + self.restrained_ligand_atoms
# Set spring constants uniformly, as in Ref [1] Table 1 caption.
_assign_if_undefined('K_r', 20.0 * unit.kilocalories_per_mole / unit.angstrom**2)
for parameter_name in ['K_thetaA', 'K_thetaB', 'K_phiA', 'K_phiB', 'K_phiC']:
_assign_if_undefined(parameter_name, 20.0 * unit.kilocalories_per_mole / unit.radian**2)
# Measure equilibrium geometries from static reference structure
t = md.Trajectory(sampler_states.positions / unit.nanometers, topography.topology)
atom_pairs = [restrained_atoms[2:4]]
distances = md.geometry.compute_distances(t, atom_pairs, periodic=False)
_assign_if_undefined('r_aA0', distances[0][0] * unit.nanometers)
atom_triplets = [restrained_atoms[i:(i+3)] for i in range(1, 3)]
angles = md.geometry.compute_angles(t, atom_triplets, periodic=False)
for parameter_name, angle in zip(['theta_A0', 'theta_B0'], angles[0]):
_assign_if_undefined(parameter_name, angle * unit.radians)
atom_quadruplets = [restrained_atoms[i:(i+4)] for i in range(3)]
dihedrals = md.geometry.compute_dihedrals(t, atom_quadruplets, periodic=False)
for parameter_name, angle in zip(['phi_A0', 'phi_B0', 'phi_C0'], dihedrals[0]):
_assign_if_undefined(parameter_name, angle * unit.radians)
# Write restraint parameters
msg = 'restraint parameters:\n'
for parameter_name, parameter_value in self._parameters.items():
msg += '%24s : %s\n' % (parameter_name, parameter_value)
logger.debug(msg)
if __name__ == '__main__':
import doctest
doctest.testmod()
# doctest.run_docstring_examples(Harmonic, globals())