from __future__ import annotations
from typing import Optional, Union
from pharmpy.basic import Expr, TExpr, TSymbol
from pharmpy.model import Assignment, Model, Parameter, Parameters
from .expressions import _create_symbol
from .odes import find_clearance_parameters, find_volume_parameters
[docs]
def add_allometry(
model: Model,
allometric_variable: Optional[TSymbol] = None,
reference_value: TExpr = 70,
parameters: Optional[list[TExpr]] = None,
initials: Optional[list[Union[int, float]]] = None,
lower_bounds: Optional[list[Union[int, float]]] = None,
upper_bounds: Optional[list[Union[int, float]]] = None,
fixed: bool = True,
):
"""Add allometric scaling of parameters
Add an allometric function to each listed parameter. The function will be
P=P*(X/Z)**T where P is the parameter, X the allometric_variable, Z the reference_value
and T is a theta. Default is to automatically use clearance and volume parameters.
If there already exists a covariate effect (or allometric scaling) on a parameter
with the specified allometric variable, nothing will be added.
If no allometric variable is specified, it will be extracted from the dataset based on
the descriptor "body weight".
Parameters
----------
model : Model
Pharmpy model
allometric_variable : str or TExpr
Value to use for allometry (X above)
reference_value : str, int, float or TExpr
Reference value (Z above)
parameters : list
Parameters to use or None (default) for all available CL, Q and V parameters
initials : list
Initial estimates for the exponents. Default is to use 0.75 for CL and Qs and 1 for Vs
lower_bounds : list
Lower bounds for the exponents. Default is 0 for all parameters
upper_bounds : list
Upper bounds for the exponents. Default is 2 for all parameters
fixed : bool
Whether the exponents should be fixed
Return
------
Model
Updated Pharmpy model
Examples
--------
>>> from pharmpy.modeling import load_example_model, add_allometry, remove_covariate_effect
>>> model = load_example_model("pheno")
>>> model = remove_covariate_effect(model, 'CL', 'WGT')
>>> model = remove_covariate_effect(model, 'V', 'WGT')
>>> model = add_allometry(model, allometric_variable='WGT')
>>> model.statements.before_odes
TVCL = POP_CL
TVV = POP_VC
⎧TVV⋅(COVAPGR + 1) for APGR < 5
⎨
TVV = ⎩ TVV otherwise
ETA_CL
CL = TVCL⋅ℯ
ALLO_CL
⎛WGT⎞
CL⋅⎜───⎟
CL = ⎝70 ⎠
ETA_VC
VC = TVV⋅ℯ
ALLO_VC
⎛WGT⎞
VC⋅⎜───⎟
VC = ⎝70 ⎠
V = VC
S₁ = VC
"""
from pharmpy.modeling import has_covariate_effect
if allometric_variable is None:
try:
allometric_variable = model.datainfo.descriptorix["body weight"][0].name
except IndexError:
raise ValueError(
"No allometric variable could be found. Try setting the allometric_variable argument"
)
variable = Expr(allometric_variable)
reference = Expr(reference_value)
if parameters is not None:
parsed_parameters = [Expr(p) for p in parameters]
else:
parsed_parameters = []
if parameters is None or initials is None:
cls = find_clearance_parameters(model)
vcs = find_volume_parameters(model)
if parameters is None:
parsed_parameters = cls + vcs
if initials is None:
# Need to understand which parameter is CL or Q and which is V
initials = []
for p in parsed_parameters:
if p in cls:
initials.append(0.75)
elif p in vcs:
initials.append(1.0)
else:
raise ValueError(
f"Parameter {p} could not be mapped to CL/Q or V: argument `initials` must be provided"
)
if not parsed_parameters:
raise ValueError("No parameters provided")
if lower_bounds is None:
lower_bounds = [0.0] * len(parsed_parameters)
if upper_bounds is None:
upper_bounds = [2.0] * len(parsed_parameters)
if not (len(parsed_parameters) == len(initials) == len(lower_bounds) == len(upper_bounds)):
raise ValueError("The number of parameters, initials and bounds must be the same")
sset = model.statements
params = list(model.parameters)
for p, init, lower, upper in zip(parsed_parameters, initials, lower_bounds, upper_bounds):
if not has_covariate_effect(model, str(p), str(variable)):
symb = _create_symbol(
sset, params, model.random_variables, model.datainfo, f'ALLO_{p.name}', False
)
param = Parameter(symb.name, init=init, lower=lower, upper=upper, fix=fixed)
params.append(param)
expr = p * (variable / reference) ** param.symbol
new_ass = Assignment.create(p, expr)
ind = sset.find_assignment_index(p)
assert ind is not None
sset = sset[0 : ind + 1] + new_ass + sset[ind + 1 :]
model = model.replace(statements=sset, parameters=Parameters.create(params))
model = model.update_source()
return model
