"""
A collection of protocols for reweighting cached simulation data.
"""
import abc
import typing
from os import path
import numpy as np
import pint
import pymbar
from scipy.special import logsumexp
from openff.evaluator import unit
from openff.evaluator.attributes import UNDEFINED
from openff.evaluator.thermodynamics import ThermodynamicState
from openff.evaluator.utils.statistics import ObservableType, StatisticsArray, bootstrap
from openff.evaluator.workflow import Protocol, workflow_protocol
from openff.evaluator.workflow.attributes import InputAttribute, OutputAttribute
[docs]@workflow_protocol()
class ConcatenateTrajectories(Protocol):
"""A protocol which concatenates multiple trajectories into
a single one.
"""
input_coordinate_paths = InputAttribute(
docstring="A list of paths to the starting PDB coordinates for each of the trajectories.",
type_hint=list,
default_value=UNDEFINED,
)
input_trajectory_paths = InputAttribute(
docstring="A list of paths to the trajectories to concatenate.",
type_hint=list,
default_value=UNDEFINED,
)
output_coordinate_path = OutputAttribute(
docstring="The path the PDB coordinate file which contains the topology "
"of the concatenated trajectory.",
type_hint=str,
)
output_trajectory_path = OutputAttribute(
docstring="The path to the concatenated trajectory.", type_hint=str
)
def _execute(self, directory, available_resources):
import mdtraj
if len(self.input_coordinate_paths) != len(self.input_trajectory_paths):
raise ValueError(
"There should be the same number of coordinate and trajectory paths."
)
if len(self.input_trajectory_paths) == 0:
raise ValueError("No trajectories were given to concatenate.")
trajectories = []
output_coordinate_path = None
for coordinate_path, trajectory_path in zip(
self.input_coordinate_paths, self.input_trajectory_paths
):
output_coordinate_path = output_coordinate_path or coordinate_path
trajectories.append(mdtraj.load_dcd(trajectory_path, coordinate_path))
self.output_coordinate_path = output_coordinate_path
output_trajectory = (
trajectories[0]
if len(trajectories) == 1
else mdtraj.join(trajectories, False, False)
)
self.output_trajectory_path = path.join(directory, "output_trajectory.dcd")
output_trajectory.save_dcd(self.output_trajectory_path)
[docs]@workflow_protocol()
class ConcatenateStatistics(Protocol):
"""A protocol which concatenates multiple trajectories into
a single one.
"""
input_statistics_paths = InputAttribute(
docstring="A list of paths to statistics arrays to concatenate.",
type_hint=list,
default_value=UNDEFINED,
)
output_statistics_path = OutputAttribute(
docstring="The path the csv file which contains the concatenated statistics.",
type_hint=str,
)
def _execute(self, directory, available_resources):
if len(self.input_statistics_paths) == 0:
raise ValueError("No statistics arrays were given to concatenate.")
arrays = [
StatisticsArray.from_pandas_csv(file_path)
for file_path in self.input_statistics_paths
]
if len(arrays) > 1:
output_array = StatisticsArray.join(*arrays)
else:
output_array = arrays[0]
self.output_statistics_path = path.join(directory, "output_statistics.csv")
output_array.to_pandas_csv(self.output_statistics_path)
[docs]@workflow_protocol()
class BaseReducedPotentials(Protocol, abc.ABC):
"""A base class for protocols which will re-evaluate the reduced potential
of a series of configurations for a given set of force field parameters.
"""
thermodynamic_state = InputAttribute(
docstring="The state to calculate the reduced potentials at.",
type_hint=ThermodynamicState,
default_value=UNDEFINED,
)
system_path = InputAttribute(
docstring="The path to the system object which describes the systems "
"potential energy function.",
type_hint=str,
default_value=UNDEFINED,
)
enable_pbc = InputAttribute(
docstring="If true, periodic boundary conditions will be enabled.",
type_hint=bool,
default_value=True,
)
coordinate_file_path = InputAttribute(
docstring="The path to the coordinate file which contains topology "
"information about the system.",
type_hint=str,
default_value=UNDEFINED,
)
trajectory_file_path = InputAttribute(
docstring="The path to the trajectory file which contains the "
"configurations to calculate the energies of.",
type_hint=str,
default_value=UNDEFINED,
)
kinetic_energies_path = InputAttribute(
docstring="The file path to a statistics array which contain the kinetic "
"energies of each frame in the trajectory.",
type_hint=str,
default_value=UNDEFINED,
)
high_precision = InputAttribute(
docstring="If true, the reduced potentials will be calculated using double "
"precision operations.",
type_hint=bool,
default_value=False,
)
use_internal_energy = InputAttribute(
docstring="If true the internal energy, rather than the potential energy will "
"be used when calculating the reduced potential. This is required "
"when reweighting properties which depend on the total energy, such "
"as enthalpy.",
type_hint=bool,
default_value=False,
)
statistics_file_path = OutputAttribute(
docstring="A file path to the statistics file which contains the reduced "
"potentials, and the potential, kinetic and total energies and "
"enthalpies evaluated at the specified state and using the "
"specified system object.",
type_hint=str,
)
[docs]class BaseMBARProtocol(Protocol, abc.ABC):
"""Reweights a set of observables using MBAR to calculate
the average value of the observables at a different state
than they were originally measured.
"""
reference_reduced_potentials = InputAttribute(
docstring="A list of paths to the reduced potentials of each "
"reference state.",
type_hint=typing.Union[str, list],
default_value=UNDEFINED,
)
target_reduced_potentials = InputAttribute(
docstring="A list of paths to the reduced potentials of the target state.",
type_hint=typing.Union[str, list],
default_value=UNDEFINED,
)
bootstrap_uncertainties = InputAttribute(
docstring="If true, bootstrapping will be used to estimated the total uncertainty",
type_hint=bool,
default_value=False,
)
bootstrap_iterations = InputAttribute(
docstring="The number of bootstrap iterations to perform if bootstraped "
"uncertainties have been requested",
type_hint=int,
default_value=1,
)
bootstrap_sample_size = InputAttribute(
docstring="The relative bootstrap sample size to use if bootstraped "
"uncertainties have been requested",
type_hint=float,
default_value=1.0,
)
required_effective_samples = InputAttribute(
docstring="The minimum number of MBAR effective samples for the "
"reweighted value to be trusted. If this minimum is not met "
"then the uncertainty will be set to sys.float_info.max",
type_hint=int,
default_value=50,
)
value = OutputAttribute(
docstring="The reweighted average value of the observable at the target state.",
type_hint=pint.Measurement,
)
effective_samples = OutputAttribute(
docstring="The number of effective samples which were reweighted.",
type_hint=float,
)
effective_sample_indices = OutputAttribute(
docstring="The indices of those samples which have a non-zero weight.",
type_hint=list,
)
[docs] def __init__(self, protocol_id):
super().__init__(protocol_id)
self._reference_observables = []
def _execute(self, directory, available_resources):
if len(self._reference_observables) == 0:
raise ValueError("There were no observables to reweight.")
if not isinstance(self._reference_observables[0], pint.Quantity):
raise ValueError(
"The reference_observables input should be a list of pint.Quantity "
"wrapped ndarray's.",
)
observables = self._prepare_observables_array(self._reference_observables)
observable_unit = self._reference_observables[0].units
if self.bootstrap_uncertainties:
self._execute_with_bootstrapping(observable_unit, observables=observables)
else:
self._execute_without_bootstrapping(
observable_unit, observables=observables
)
def _load_reduced_potentials(self):
"""Loads the target and reference reduced potentials
from the specified statistics files.
Returns
-------
numpy.ndarray
The reference reduced potentials array with dtype=double and
shape=(1,)
numpy.ndarray
The target reduced potentials array with dtype=double and
shape=(1,)
"""
if isinstance(self.reference_reduced_potentials, str):
self.reference_reduced_potentials = [self.reference_reduced_potentials]
if isinstance(self.target_reduced_potentials, str):
self.target_reduced_potentials = [self.target_reduced_potentials]
reference_reduced_potentials = []
target_reduced_potentials = []
# Load in the reference reduced potentials.
for file_path in self.reference_reduced_potentials:
statistics_array = StatisticsArray.from_pandas_csv(file_path)
reduced_potentials = statistics_array[ObservableType.ReducedPotential]
reference_reduced_potentials.append(
reduced_potentials.to(unit.dimensionless).magnitude
)
# Load in the target reduced potentials.
if len(target_reduced_potentials) > 1:
raise ValueError(
"This protocol currently only supports reweighting to "
"a single target state."
)
for file_path in self.target_reduced_potentials:
statistics_array = StatisticsArray.from_pandas_csv(file_path)
reduced_potentials = statistics_array[ObservableType.ReducedPotential]
target_reduced_potentials.append(
reduced_potentials.to(unit.dimensionless).magnitude
)
reference_reduced_potentials = np.array(reference_reduced_potentials)
target_reduced_potentials = np.array(target_reduced_potentials)
return reference_reduced_potentials, target_reduced_potentials
def _execute_with_bootstrapping(self, observable_unit, **observables):
"""Calculates the average reweighted observables at the target state,
using bootstrapping to estimate uncertainties.
Parameters
----------
observable_unit: pint.Unit:
The expected unit of the reweighted observable.
observables: dict of str and numpy.ndarray
The observables to reweight which have been stripped of their units.
"""
(
reference_reduced_potentials,
target_reduced_potentials,
) = self._load_reduced_potentials()
frame_counts = np.array(
[len(observable) for observable in self._reference_observables]
)
# Construct a dummy mbar object to get out the number of effective samples.
mbar = self._construct_mbar_object(reference_reduced_potentials)
(
self.effective_samples,
effective_sample_indices,
) = self._compute_effective_samples(mbar, target_reduced_potentials)
if self.effective_samples < self.required_effective_samples:
raise ValueError(
f"There was not enough effective samples to reweight - "
f"{self.effective_samples} < {self.required_effective_samples}"
)
# Transpose the observables ready for bootstrapping.
reference_reduced_potentials = np.transpose(reference_reduced_potentials)
target_reduced_potentials = np.transpose(target_reduced_potentials)
transposed_observables = {}
for observable_key in observables:
transposed_observables[observable_key] = np.transpose(
observables[observable_key]
)
value, uncertainty = bootstrap(
self._bootstrap_function,
self.bootstrap_iterations,
self.bootstrap_sample_size,
frame_counts,
reference_reduced_potentials=reference_reduced_potentials,
target_reduced_potentials=target_reduced_potentials,
**transposed_observables,
)
self.effective_sample_indices = effective_sample_indices
self.value = (value * observable_unit).plus_minus(uncertainty * observable_unit)
def _execute_without_bootstrapping(self, observable_unit, **observables):
"""Calculates the average reweighted observables at the target state,
using the built-in pymbar method to estimate uncertainties.
Parameters
----------
observables: dict of str and numpy.ndarray
The observables to reweight which have been stripped of their units.
"""
if len(observables) > 1:
raise ValueError(
"Currently only a single observable can be reweighted at"
"any one time."
)
(
reference_reduced_potentials,
target_reduced_potentials,
) = self._load_reduced_potentials()
values, uncertainties, self.effective_samples = self._reweight_observables(
reference_reduced_potentials, target_reduced_potentials, **observables
)
observable_key = next(iter(observables))
uncertainty = uncertainties[observable_key]
if self.effective_samples < self.required_effective_samples:
raise ValueError(
f"There was not enough effective samples to reweight - "
f"{self.effective_samples} < {self.required_effective_samples}"
)
self.value = (values[observable_key] * observable_unit).plus_minus(
uncertainty * observable_unit
)
@staticmethod
def _prepare_observables_array(reference_observables):
"""Takes a list of reference observables, and concatenates them
into a single `pint.Quantity` wrapped numpy array.
Parameters
----------
reference_observables: List of pint.Quantity
A list of observables for each reference state,
which each observable is a `pint.Quantity` wrapped numpy
array.
Returns
-------
np.ndarray
A unitless numpy array of all of the observables.
"""
frame_counts = np.array(
[len(observable) for observable in reference_observables]
)
number_of_configurations = frame_counts.sum()
observable_dimensions = (
1
if len(reference_observables[0].shape) == 1
else reference_observables[0].shape[1]
)
observable_unit = reference_observables[0].units
observables = np.zeros((observable_dimensions, number_of_configurations))
# Build up an array which contains the observables from all
# of the reference states.
for index_k, observables_k in enumerate(reference_observables):
start_index = np.array(frame_counts[0:index_k]).sum()
for index in range(0, frame_counts[index_k]):
value = observables_k[index].to(observable_unit).magnitude
if not isinstance(value, np.ndarray):
observables[0][start_index + index] = value
continue
for dimension in range(observable_dimensions):
observables[dimension][start_index + index] = value[dimension]
return observables
def _bootstrap_function(
self,
reference_reduced_potentials,
target_reduced_potentials,
**reference_observables,
):
"""The function which will be called after each bootstrap
iteration, if bootstrapping is being employed to estimated
the reweighting uncertainty.
Parameters
----------
reference_reduced_potentials
target_reduced_potentials
reference_observables
Returns
-------
float
The bootstrapped value,
"""
assert len(reference_observables) == 1
transposed_observables = {}
for key in reference_observables:
transposed_observables[key] = np.transpose(reference_observables[key])
values, _, _ = self._reweight_observables(
np.transpose(reference_reduced_potentials),
np.transpose(target_reduced_potentials),
**transposed_observables,
)
return next(iter(values.values()))
def _construct_mbar_object(self, reference_reduced_potentials):
"""Constructs a new `pymbar.MBAR` object for a given set of reference
and target reduced potentials
Parameters
-------
reference_reduced_potentials: numpy.ndarray
The reference reduced potentials.
Returns
-------
pymbar.MBAR
The constructed `MBAR` object.
"""
frame_counts = np.array(
[len(observables) for observables in self._reference_observables]
)
# Construct the mbar object.
mbar = pymbar.MBAR(
reference_reduced_potentials,
frame_counts,
verbose=False,
relative_tolerance=1e-12,
)
return mbar
@staticmethod
def _compute_effective_samples(mbar, target_reduced_potentials):
"""Compute the effective number of samples which contribute to the final
reweighted estimate.
Parameters
----------
mbar: pymbar.MBAR
The MBAR object which contains the sample weights.
target_reduced_potentials: numpy.ndarray
The target reduced potentials.
Returns
-------
int
The effective number of samples.
list of int
The indices of samples which have non-zero weights.
"""
states_with_samples = mbar.N_k > 0
log_ref_q_k = mbar.f_k[states_with_samples] - mbar.u_kn[states_with_samples].T
log_denominator_n = logsumexp(
log_ref_q_k, b=mbar.N_k[states_with_samples], axis=1
)
target_f_hat = -logsumexp(
-target_reduced_potentials[: len(target_reduced_potentials)]
- log_denominator_n,
axis=1,
)
log_tar_q_k = target_f_hat - target_reduced_potentials
# Calculate the weights
weights = np.exp(log_tar_q_k - log_denominator_n)
effective_samples = 1.0 / np.sum(weights ** 2)
effective_sample_indices = [
index
for index in range(weights.shape[1])
if not np.isclose(weights[0][index], 0.0)
]
return effective_samples, effective_sample_indices
def _reweight_observables(
self,
reference_reduced_potentials,
target_reduced_potentials,
**reference_observables,
):
"""Reweights a set of reference observables to
the target state.
Returns
-------
dict of str and float or list of float
The reweighted values.
dict of str and float or list of float
The MBAR calculated uncertainties in the reweighted values.
int
The number of effective samples.
"""
# Construct the mbar object.
mbar = self._construct_mbar_object(reference_reduced_potentials)
(
effective_samples,
self.effective_sample_indices,
) = self._compute_effective_samples(mbar, target_reduced_potentials)
values = {}
uncertainties = {}
for observable_key in reference_observables:
reference_observable = reference_observables[observable_key]
observable_dimensions = reference_observable.shape[0]
values[observable_key] = np.zeros((observable_dimensions, 1))
uncertainties[observable_key] = np.zeros((observable_dimensions, 1))
for dimension in range(observable_dimensions):
results = mbar.computeExpectations(
reference_observable[dimension],
target_reduced_potentials,
state_dependent=True,
)
values[observable_key][dimension] = results[0][-1]
uncertainties[observable_key][dimension] = results[1][-1]
if observable_dimensions == 1:
values[observable_key] = values[observable_key][0][0].item()
uncertainties[observable_key] = uncertainties[observable_key][0][
0
].item()
return values, uncertainties, effective_samples
[docs]@workflow_protocol()
class ReweightStatistics(BaseMBARProtocol):
"""Reweights a set of observables from a `StatisticsArray` using MBAR.
"""
statistics_paths = InputAttribute(
docstring="The file paths to the statistics array which contains the observables "
"of interest from each state. If the observable of interest is "
"dependant on the changing variable (e.g. the potential energy) then "
"this must be a path to the observable re-evaluated at the new state.",
type_hint=typing.Union[list, str],
default_value=UNDEFINED,
)
statistics_type = InputAttribute(
docstring="The type of observable to reweight.",
type_hint=ObservableType,
default_value=UNDEFINED,
)
frame_counts = InputAttribute(
docstring="A list which describes how many of the statistics in the array "
"belong to each reference state. If this input is used, only a single file "
"path should be passed to the `statistics_paths` input.",
type_hint=list,
default_value=[],
optional=True,
)
def _execute(self, directory, available_resources):
if isinstance(self.statistics_paths, str):
self.statistics_paths = [self.statistics_paths]
if self.statistics_paths is None or len(self.statistics_paths) == 0:
return ValueError("No statistics paths were provided.")
if len(self.frame_counts) > 0 and len(self.statistics_paths) != 1:
raise ValueError(
"The frame counts input can only be used when only a single "
"path is passed to the `statistics_paths` input.",
)
if self.statistics_type == ObservableType.KineticEnergy:
raise ValueError("Kinetic energies cannot be reweighted.")
statistics_arrays = [
StatisticsArray.from_pandas_csv(file_path)
for file_path in self.statistics_paths
]
self._reference_observables = []
if len(self.frame_counts) > 0:
statistics_array = statistics_arrays[0]
current_index = 0
for frame_count in self.frame_counts:
if frame_count <= 0:
raise ValueError("The frame counts must be > 0.")
observables = statistics_array[self.statistics_type][
current_index : current_index + frame_count
]
self._reference_observables.append(observables)
current_index += frame_count
else:
for statistics_array in statistics_arrays:
observables = statistics_array[self.statistics_type]
self._reference_observables.append(observables)
return super(ReweightStatistics, self)._execute(directory, available_resources)