def _set_initial_concentration(condition_id, species_id,
init_par_id,
par_map, scale_map):
value = petab.to_float_if_float(
self.petab_problem.condition_df.loc[
condition_id, species_id])
if isinstance(value, float):
# numeric initial state
par_map[init_par_id] = value
scale_map[init_par_id] = ptc.LIN
else:
# parametric initial state
try:
# try find in mapping
par_map[init_par_id] = par_map[value]
scale_map[init_par_id] = scale_map[value]
except KeyError:
# otherwise look up in parameter table
if (self.petab_problem.parameter_df.loc[
value, ptc.ESTIMATE] == 0):
par_map[init_par_id] = \
self.petab_problem.parameter_df.loc[
value, ptc.NOMINAL_VALUE]
else:
par_map[init_par_id] = value
if (ptc.PARAMETER_SCALE
not in self.petab_problem.parameter_df
or not self.petab_problem.parameter_df.loc[
value, ptc.PARAMETER_SCALE]):
scale_map[init_par_id] = ptc.LIN
else:
scale_map[init_par_id] = \
self.petab_problem.parameter_df.loc[
value, ptc.PARAMETER_SCALE]
def write_measurements(self):
"""
Write measurements to hdf5 dataset
"""
# create inverse mapping for faster lookup
observable_id_to_index = {
name: idx for idx, name in enumerate(self.observable_ids)}
measurement_df = self.petab_problem.measurement_df
if ptc.NOISE_PARAMETERS not in measurement_df:
measurement_df[ptc.NOISE_PARAMETERS] = np.nan
for sim_idx, (preeq_cond_idx, sim_cond_idx) \
in enumerate(self.condition_map):
# print("Condition", sim_idx, (preeq_cond_idx, sim_cond_idx))
cur_mes_df = self._get_measurements_for_condition(
sim_idx, preeq_cond_idx, sim_cond_idx)
# get the required, possibly non-unique (replicates) timepoints
grp = cur_mes_df.groupby([ptc.OBSERVABLE_ID, ptc.TIME]).size() \
.reset_index().rename(columns={0: 'sum'}) \
.groupby([ptc.TIME])['sum'].agg(['max']).reset_index()
timepoints = np.sort(np.repeat(grp[ptc.TIME], grp['max'])).tolist()
mes = np.full(shape=(len(timepoints), self.ny), fill_value=np.nan)
sd = mes.copy()
# write measurements from each row in measurementDf
for index, row in cur_mes_df.iterrows():
observable_idx = observable_id_to_index[row[ptc.OBSERVABLE_ID]]
time_idx = timepoints.index(row.time)
while not np.isnan(mes[time_idx, observable_idx]):
time_idx += 1
if timepoints[time_idx] != row.time:
raise AssertionError(
"Replicate handling failed for "
f'{row[ptc.SIMULATION_CONDITION_ID]} - '
f'{row[ptc.OBSERVABLE_ID]}'
f' time {row[ptc.TIME]}\n' + str(cur_mes_df))
mes[time_idx, observable_idx] = float(row[ptc.MEASUREMENT])
sigma = to_float_if_float(row[ptc.NOISE_PARAMETERS])
if isinstance(sigma, float):
sd[time_idx, observable_idx] = sigma
# write to file
g = self.f.require_group("measurements")
g.create_dataset(name=f"y/{sim_idx}", data=mes, dtype='f8',
compression=self.compression)
g.create_dataset(name=f"ysigma/{sim_idx}", data=sd, dtype='f8',
compression=self.compression)
g.create_dataset(name=f"t/{sim_idx}", data=timepoints, dtype='f8',
compression=self.compression)
def _set_initial_concentration(condition_id, species_id, init_par_id,
par_map, scale_map):
value = petab.to_float_if_float(
petab_problem.condition_df.loc[condition_id, species_id])
par_map[init_par_id] = value
if isinstance(value, float):
# numeric initial state
scale_map[init_par_id] = petab.LIN
else:
# parametric initial state
scale_map[init_par_id] = petab_problem.parameter_df.loc[
value, PARAMETER_SCALE]
def _set_initial_concentration(condition_id, species_id, init_par_id,
par_map, scale_map):
value = petab.to_float_if_float(
petab_problem.condition_df.loc[condition_id, species_id])
if pd.isna(value):
value = float(
get_species_initial(
petab_problem.sbml_model.getSpecies(species_id)))
logger.debug(
f'The species {species_id} has no initial value '
f'defined for the condition {condition_id} in '
'the PEtab conditions table. The initial value is '
f'now set to {value}, which is the initial value '
'defined in the SBML model.')
par_map[init_par_id] = value
if isinstance(value, float):
# numeric initial state
scale_map[init_par_id] = petab.LIN
else:
# parametric initial state
scale_map[init_par_id] = petab_problem.parameter_df.loc[
value, PARAMETER_SCALE]
def _generate_simulation_to_optimization_parameter_mapping(self) -> None:
"""
Create dataset n_parameters_simulation x n_conditions with indices of
respective parameters in parameters_optimization
"""
# get list of tuple of parameters dicts for all conditions
self.parameter_mapping = self.petab_problem \
.get_optimization_to_simulation_parameter_mapping(
warn_unmapped=False, scaled_parameters=False)
variable_par_ids = self.amici_model.getParameterIds()
fixed_par_ids = self.amici_model.getFixedParameterIds()
# Translate parameter ID mapping to index mapping
# create inverse mapping for faster lookup
print(self.problem_parameter_ids)
optimization_parameter_name_to_index = {
name: idx for idx, name in enumerate(self.problem_parameter_ids)}
self.optimization_parameter_name_to_index = \
optimization_parameter_name_to_index
# use in-memory matrix, don't write every entry to file directly
num_model_parameters = self.amici_model.np()
mapping_matrix = np.zeros(
shape=(num_model_parameters, self.condition_map.shape[0]),
dtype='<i4')
override_matrix = np.full(shape=mapping_matrix.shape,
fill_value=np.nan)
pscale_matrix = np.full(shape=mapping_matrix.shape, dtype='<i4',
fill_value=amici.ParameterScaling_none)
# AMICI fixed parameters
self.nk = len(fixed_par_ids)
print(Fore.CYAN + "Number of fixed parameters:",
len(fixed_par_ids))
# TODO: can fixed parameter differ between same condition vector used
# in different sim x preeq combinations?
fixed_parameter_matrix = np.full(
shape=(self.nk, self.num_condition_vectors),
fill_value=np.nan)
# For handling initial states
species_in_condition_table = [
col for col in self.petab_problem.condition_df
if self.petab_problem.sbml_model.getSpecies(col) is not None]
# Merge and preeq and sim parameters, filter fixed parameters
for condition_idx, \
(condition_map_preeq, condition_map_sim,
condition_scale_map_preeq, condition_scale_map_sim) \
in enumerate(self.parameter_mapping):
preeq_cond_idx, sim_cond_idx = self.condition_map[condition_idx]
preeq_cond_id = self.condition_ids[preeq_cond_idx] \
if preeq_cond_idx != NO_PREEQ_CONDITION_IDX else None
sim_cond_id = self.condition_ids[sim_cond_idx]
print(preeq_cond_idx, preeq_cond_id, sim_cond_idx, sim_cond_id)
if len(condition_map_preeq) != len(condition_scale_map_preeq) \
or len(condition_map_sim) != len(condition_scale_map_sim):
raise AssertionError(
"Number of parameters and number of parameter "
"scales do not match.")
if len(condition_map_preeq) \
and len(condition_map_preeq) != len(condition_map_sim):
# logger.debug(
# f"Preequilibration parameter map: {condition_map_preeq}")
# logger.debug(f"Simulation parameter map: {condition_map_sim}")
raise AssertionError(
"Number of parameters for preequilibration "
"and simulation do not match.")
# TODO: requires special handling of initial concentrations
if species_in_condition_table:
# set indicator fixed parameter for preeq
# (we expect here, that this parameter was added during AMICI
# model import and that it was not added by the user with a
# different meaning...)
if preeq_cond_idx != NO_PREEQ_CONDITION_IDX:
condition_map_preeq[PREEQ_INDICATOR_ID] = 1.0
condition_scale_map_preeq[PREEQ_INDICATOR_ID] = ptc.LIN
condition_map_sim[PREEQ_INDICATOR_ID] = 0.0
condition_scale_map_sim[PREEQ_INDICATOR_ID] = ptc.LIN
def _set_initial_concentration(condition_id, species_id,
init_par_id,
par_map, scale_map):
value = petab.to_float_if_float(
self.petab_problem.condition_df.loc[
condition_id, species_id])
if isinstance(value, float):
# numeric initial state
par_map[init_par_id] = value
scale_map[init_par_id] = ptc.LIN
else:
# parametric initial state
try:
# try find in mapping
par_map[init_par_id] = par_map[value]
scale_map[init_par_id] = scale_map[value]
except KeyError:
# otherwise look up in parameter table
if (self.petab_problem.parameter_df.loc[
value, ptc.ESTIMATE] == 0):
par_map[init_par_id] = \
self.petab_problem.parameter_df.loc[
value, ptc.NOMINAL_VALUE]
else:
par_map[init_par_id] = value
if (ptc.PARAMETER_SCALE
not in self.petab_problem.parameter_df
or not self.petab_problem.parameter_df.loc[
value, ptc.PARAMETER_SCALE]):
scale_map[init_par_id] = ptc.LIN
else:
scale_map[init_par_id] = \
self.petab_problem.parameter_df.loc[
value, ptc.PARAMETER_SCALE]
for species_id in species_in_condition_table:
# for preequilibration
init_par_id = f'initial_{species_id}_preeq'
# need to set dummy value for preeq parameter anyways, as it
# is expected below (set to 0, not nan, because will be
# multiplied with indicator variable in initial assignment)
condition_map_sim[init_par_id] = 0.0
condition_scale_map_sim[init_par_id] = ptc.LIN
if preeq_cond_idx != NO_PREEQ_CONDITION_IDX:
_set_initial_concentration(
preeq_cond_id, species_id, init_par_id,
condition_map_preeq,
condition_scale_map_preeq)
# enable state reinitialization
self.f['/fixedParameters/simulationConditions'][condition_idx, 2] = 1
# for simulation
init_par_id = f'initial_{species_id}_sim'
_set_initial_concentration(
sim_cond_id, species_id, init_par_id,
condition_map_sim,
condition_scale_map_sim)
print(condition_map_preeq, condition_map_sim)
# split into fixed and variable parameters:
condition_map_preeq_var = condition_scale_map_preeq_var = None
condition_map_preeq_fix = None
if condition_map_preeq:
condition_map_preeq_var, condition_map_preeq_fix = \
subset_dict(condition_map_preeq, variable_par_ids,
fixed_par_ids)
condition_scale_map_preeq_var, _ = \
subset_dict(condition_scale_map_preeq, variable_par_ids,
fixed_par_ids)
condition_map_sim_var, condition_map_sim_fix = \
subset_dict(condition_map_sim, variable_par_ids, fixed_par_ids)
condition_scale_map_sim_var, condition_scale_map_sim_fix = \
subset_dict(condition_scale_map_sim, variable_par_ids,
fixed_par_ids)
if condition_map_preeq:
# merge after having removed potentially fixed parameters
# which may differ between preequilibration and simulation
petab.merge_preeq_and_sim_pars_condition(
condition_map_preeq_var, condition_map_sim_var,
condition_scale_map_preeq_var, condition_scale_map_sim_var,
condition_idx)
print(self.problem_parameter_ids)
# mapping for each model parameter
for model_parameter_idx, model_parameter_id \
in enumerate(variable_par_ids):
mapped_parameter = condition_map_sim[model_parameter_id]
mapped_parameter = to_float_if_float(mapped_parameter)
try:
(mapped_idx, override) = self.get_index_mapping_for_par(
mapped_parameter, optimization_parameter_name_to_index)
mapping_matrix[model_parameter_idx, condition_idx] = \
mapped_idx
override_matrix[model_parameter_idx, condition_idx] = \
override
except IndexError as e:
print(Fore.RED + "Error in parameter mapping:", e)
print(model_parameter_idx, mapped_parameter)
print(self.parameter_mapping)
raise e
# Set parameter scales for simulation
pscale_matrix[:, condition_idx] = list(
petab_scale_to_amici_scale(condition_scale_map_sim[amici_id])
for amici_id in variable_par_ids)
fixed_parameter_matrix[:, self.condition_map[condition_idx, 1]] = \
np.array([condition_map_sim_fix[par_id]
for par_id in fixed_par_ids])
if condition_map_preeq:
fixed_parameter_matrix[:, self.condition_map[condition_idx, 0]] = \
np.array([condition_map_preeq_fix[par_id]
for par_id in fixed_par_ids])
# write to file
self.create_fixed_parameter_dataset_and_write_attributes(
fixed_par_ids, fixed_parameter_matrix)
self.f.require_dataset('/parameters/pscaleSimulation',
shape=pscale_matrix.T.shape, dtype="<i4",
data=pscale_matrix.T)
write_parameter_map(self.f, mapping_matrix, override_matrix,
num_model_parameters, self.compression)
# for cost function parameters
petab_opt_par_scale = \
self.petab_problem.get_optimization_parameter_scales()
pscale_opt_par = np.array(
list(petab_scale_to_amici_scale(petab_opt_par_scale[par_id])
for par_id in self.problem_parameter_ids))
self.f.require_dataset('/parameters/pscaleOptimization',
shape=pscale_opt_par.shape, dtype="<i4",
data=pscale_opt_par)
self.f.flush()
def get_model_for_condition(
petab_problem: "petab.Problem",
sim_condition_id: str = None,
preeq_condition_id: Optional[str] = None,
) -> Tuple[libsbml.SBMLDocument, libsbml.Model]:
"""Create an SBML model for the given condition.
Creates a copy of the model and updates parameters according to the PEtab
files. Estimated parameters are set to their ``nominalValue``.
Observables defined in the observables table are not added to the model.
:param petab_problem: PEtab problem
:param sim_condition_id: Simulation ``conditionId`` for which to generate a
model
:param preeq_condition_id: Preequilibration ``conditionId`` of the settings
for which to generate a model. This is only used to determine the
relevant output parameter overrides. Preequilibration is not encoded
in the resulting model.
:return: The generated SBML document, and SBML model
"""
condition_dict = {petab.SIMULATION_CONDITION_ID: sim_condition_id}
if preeq_condition_id:
condition_dict[petab.PREEQUILIBRATION_CONDITION_ID] = \
preeq_condition_id
cur_measurement_df = petab.measurements.get_rows_for_condition(
measurement_df=petab_problem.measurement_df,
condition=condition_dict,
)
parameter_map, scale_map = \
petab.parameter_mapping.get_parameter_mapping_for_condition(
condition_id=sim_condition_id,
is_preeq=False,
cur_measurement_df=cur_measurement_df,
sbml_model=petab_problem.sbml_model,
condition_df=petab_problem.condition_df,
parameter_df=petab_problem.parameter_df,
warn_unmapped=True,
scaled_parameters=False,
fill_fixed_parameters=True,
# will only become problematic once the observable and noise terms
# are added to the model
allow_timepoint_specific_numeric_noise_parameters=True,
)
# create a copy of the model
sbml_doc = petab_problem.sbml_model.getSBMLDocument().clone()
sbml_model = sbml_doc.getModel()
# fill in parameters
def get_param_value(parameter_id: str):
"""Parameter value from mapping or nominal value"""
mapped_value = parameter_map.get(parameter_id)
if not isinstance(mapped_value, str):
return mapped_value
# estimated parameter, look up in nominal parameters
return petab_problem.parameter_df.loc[mapped_value,
petab.NOMINAL_VALUE]
def remove_rules(target_id: str):
if sbml_model.removeRuleByVariable(target_id):
warn("An SBML rule was removed to set the component "
f"{target_id} to a constant value.")
sbml_model.removeInitialAssignment(target_id)
for parameter in sbml_model.getListOfParameters():
new_value = get_param_value(parameter.getId())
if new_value:
parameter.setValue(new_value)
# remove rules that would override that value
remove_rules(parameter.getId())
# set concentrations for any overridden species
for component_id in petab_problem.condition_df:
sbml_species = sbml_model.getSpecies(component_id)
if not sbml_species:
continue
# remove any rules overriding that species' initials
remove_rules(component_id)
# set initial concentration/amount
new_value = petab.to_float_if_float(
petab_problem.condition_df.loc[sim_condition_id, component_id])
if not isinstance(new_value, Number):
# parameter reference in condition table
new_value = get_param_value(new_value)
if sbml_species.isSetInitialAmount() \
or (sbml_species.getHasOnlySubstanceUnits()
and not sbml_species.isSetInitialConcentration()):
sbml_species.setInitialAmount(new_value)
else:
sbml_species.setInitialConcentration(new_value)
# set compartment size for any compartments in the condition table
for component_id in petab_problem.condition_df:
sbml_compartment = sbml_model.getCompartment(component_id)
if not sbml_compartment:
continue
# remove any rules overriding that compartment's size
remove_rules(component_id)
# set initial concentration/amount
new_value = petab.to_float_if_float(
petab_problem.condition_df.loc[sim_condition_id, component_id])
if not isinstance(new_value, Number):
# parameter reference in condition table
new_value = get_param_value(new_value)
sbml_compartment.setSize(new_value)
return sbml_doc, sbml_model