def create_data_tables(model, condition_df):
"""Create synthetic data for parameter estimation
- Simulate time-course for four different conditions
- Add gaussian noise according to selected sigma parameter
- Mimic 2 experimental batches: odd-numbered condition indices
and even-numbered conditions have different offset parameter
Arguments:
"""
timepoints = np.logspace(-5, 1, 20)
sigma_default = 0.1 # parameters are lin
sigma_parameter = 0.2
offset_batch_1 = 3.0
offset_batch_2 = 4.0
parameter_ids = list(model.getParameterIds())
observable_ids = list(model.getObservableIds())
sigma_parameter_observable_idx = \
observable_ids.index('obs_x1withsigma')
model_offset_parameter_idx = \
parameter_ids.index('observableParameter1_obs_x2_offsetted')
sigma_parameter_idx = \
parameter_ids.index('noiseParameter1_obs_x1withsigma')
# set true parameters
default_parameters = np.array(model.getParameters())
default_parameters[sigma_parameter_idx] = sigma_parameter
print('Default model parameters:')
for p, val in zip(model.getParameterIds(), model.getParameters()):
print(f'\t{p}: {val}')
print()
true_parameters = {
pid: val
for pid, val in zip(model.getParameterIds(), default_parameters)
}
# output parameters don't have default values from SBML mdoel
true_parameters['observableParameter1_obs_x1_scaled'] = 2.0
true_parameters['noiseParameter1_obs_x1withsigma'] = 0.2
true_parameters['noiseParameter1_obs_x1_scaled'] = 0.2
true_parameters['noiseParameter1_obs_x2'] = 0.2
true_parameters['noiseParameter1_obs_x1'] = 0.2
true_parameters['noiseParameter1_obs_x2_offsetted'] = 0.2
true_parameters['noiseParameter1_obs_x3'] = 0.2
true_parameters['observableParameter1_obs_x2_offsetted'] = 3.0
true_parameters['scaling_x1_common'] = \
true_parameters['observableParameter1_obs_x1_scaled']
# extend to optimization parameter vector: add second offset parameter
true_parameters['offset_x2_batch_0'] = offset_batch_1
true_parameters['offset_x2_batch_1'] = offset_batch_2
true_parameters['x1withsigma_sigma'] = sigma_parameter
print('True parameters:\t%s' % true_parameters)
# setup model
model.setTimepoints(timepoints)
model.setParameters(default_parameters)
# setup solver
solver = model.getSolver()
solver.setMaxSteps(10000)
print(condition_df)
measurement_df = petab.create_measurement_df()
print()
# set sigmas
sigmay = np.ones(shape=(model.nt(), model.nytrue)) * sigma_default
# observable with sigma parameter
sigmay[:, sigma_parameter_observable_idx] = np.nan
# llh for noisy simulated data with true parameters
expected_llh = 0.0
for condition_idx, condition_id in enumerate(condition_df.index.values):
condition_parameters = condition_df.loc[condition_id, :]
print(
f'Condition {condition_idx} "{condition_id}": {condition_parameters}'
)
# different offset for two "batches"
batch_id = condition_idx % 2
model_parameters = default_parameters[:]
if batch_id == 0:
model_parameters[model_offset_parameter_idx] = offset_batch_1
else:
model_parameters[model_offset_parameter_idx] = offset_batch_2
print('Model parameters:', model_parameters)
# simulate condition
rdata = getReturnDataForCondition(model, solver, condition_parameters,
model_parameters, sigmay,
sigma_parameter_observable_idx,
sigma_parameter_idx)
print('\tllh: ', rdata['llh'])
print('\tsllh', rdata['sllh'])
expected_llh += rdata['llh']
measurement_df = append_measurements_for_condition(
model, measurement_df, sigmay, condition_id, batch_id, rdata)
print()
print('Expected llh: ', expected_llh)
return measurement_df, true_parameters, expected_llh
def test_create_measurement_df():
"""Test measurements.create_measurement_df."""
df = petab.create_measurement_df()
assert set(df.columns.values) == set(MEASUREMENT_DF_COLS)