AMD - Drug metabolite

Will develop the best drug metabolite model based on a model or dataset input.

Regarding the dataset, two DVIDs are required.

DVID	Description
1	Assumed to be connected to the PK part of the model
2	Assumed to be connected to the metabolite part of the model

Running

The code to initiate the AMD tool for a PK model:

PythonR

from pharmpy.tools import run_amd

res = run_amd(input='path/to/dataset',
              modeltype='drug_metabolite',
              administration='iv',
              cl_init=2.0,
              vc_init=5.0,
              strategy='default',
              search_space='METABOLITE([BASIC,PSC]);PERIPHERALS(0..1,MET)',
              allometric_variable='WGT',
              occasion='VISI'
)

Arguments

The arguments used in drug metabolite models can be seen below. Some are mandatory for this type of model building while others are optional, and some AMD arguments are not used for this model type. If any of the mandatory arguments are missing, the model will not be run. The only exception being for mat_init which is only mandatory for ‘oral’ administration.

Mandatory

Argument	Description
input	Path to a dataset or start model object. See input in amd
modeltype	Set to ‘drug_metabolite’ for this model type.
cl_init	Initial estimate for the population clearance
vc_init	Initial estimate for the central compartment population volume
mat_init	Initial estimate for the mean absorption time (only for oral models)

Note

In addition to these arguments, common arguments can be added.

Strategy components

For a description about the different model building strategies in AMD, see Strategy. This section will cover the aspects that are specific to drug metabolite models.

Structural

Structural covariates

The structural covariates are added directly to the starting model. If these cannot be added here (due to missing parameters for instance) they will be added at the start of the next covsearch run. Note that all structural covariates are added all at once without any test or search.

If no structural covariates are specified, no default is used.

Modelsearch

Note

This part of the AMD tool develops the PK model, meaning it will filter out the metabolite data points. The rest of AMD uses the full dataset.

The settings that the AMD tool uses for the modelsearch subtool can be seen in the table below.

Argument	Setting
search_space	search_space (As defined in AMD options)
algorithm	‘reduced_stepwise’
iiv_strategy	‘absorption_delay’
rank_type	‘bic’ (type: mixed)
cutoff	None

If no search space is given by the user, the default search space is dependent on the administration argument

Drug metabolite ORALDrug metabolite IVDrug metabolite IV+ORAL

ABSORPTION([FO,ZO,SEQ-ZO-FO])
ELIMINATION(FO)
LAGTIME([OFF,ON])
TRANSITS([0,1,3,10],*)
PERIPHERALS([0,1])

Structsearch

For a drug metabolite model, structsearch is run to determine the best structural model. All input arguments are specified by the user when initializing AMD.

Argument	Setting
search_space	search_space (As defined in AMD options)
modeltype	‘drug_metabolite’
strictness	strictness

If no search space is given for the structsearch tool, then a default will be set to:

Drug metabolite IVDrug metabolite ORAL & drug metabolite IV+ORAL

METABOLITE([BASIC])
PERIPHERALS([0,1], MET)

IIVSearch

The settings that the AMD tool uses for this subtool can be seen in the table below.

Argument	Setting	Setting (rerun)
algorithm	‘top_down_exhaustive’	‘top_down_exhaustive’
iiv_strategy	‘fullblock’	‘no_add’
rank_type	‘mbic’ (type: iiv)	‘mbic’ (type: iiv)
cutoff	None	None
keep	Clearance parameters from input model	Clearance parameters from input model

Residual

When running the residual part of the workflow, the tool is run twice. First, the best error model is found for DVID=1` followed by the same run but for DVID=2

The settings used when running the tool can be found below. When re-running the tool, the settings remain the same.

Argument	Setting
dvid	1 (first run) and 2 (second run)
groups	4
p_value	0.05
skip	None

IOVSearch

The settings that the AMD tool uses for this subtool can be seen in the table below.

Argument	Setting
column	occasion (As defined in AMD options)
list_of_parameters	None
rank_type	‘bic’ (type: random)
cutoff	None
distribution	‘same-as-iiv’

Allometry

The settings that the AMD tool uses for this subtool can be seen in the table below.

Argument	Setting
allometric_variable	allometric_variable (As defined in AMD options)
reference_value	70
parameters	None
initials	None
lower_bounds	None
upper_bounds	None
fixed	None

covsearch

The settings that the AMD tool uses for this subtool can be seen in the table below.

Argument	Setting
search_space	search_space (As defined in AMD options)
p_forward	0.05
p_backward	0.01
max_steps	-1
algorithm	‘scm-forward-then-backward’

If no search space for this tool is given, the following default will be used:

COVARIATE?(@IIV, @CONTINUOUS, exp, *)
COVARIATE?(@IIV, @CATEGORICAL, cat, *)

Here, both statements are defined with a ‘?’, meaning that these are covariate effect(s) to be explored rather than structural covariate effects, which are added during the earlier “structural” step.

Mechanisitic covariates

If any mechanistic covariates have been given as input to the AMD tool, the specified covariate effects for these covariates is run in a separate initial covsearch run when adding covariates. These covariate effects are extracted from the given search space

Exploratory covariates

The remaining covariate effects from the search space are now run in an exploratory search.

Examples

Minimal

A minimal example for running AMD with model type PK:

PythonR

from pharmpy.tools import run_amd

dataset_path = 'path/to/dataset'

res = run_amd(
            dataset_path,
            modeltype="drug_metabolite",
            administration="iv",
            cl_init=2.0,
            vc_init=5.0
)

Model input and search space

Specifying input model and search space:

PythonR

from pharmpy.tools import run_amd

start_model = read_model('path/to/model')

res = run_amd(
            input=start_model,
            modeltype='drug_metabolite',
            search_space='ABSORPTION(FO);PERIPHERALS(1..2);METABOLITE(BASIC);PERIPHERALS(0..1,MET)',
            cl_init=2.0,
            vc_init=5.0,
            mat_init=3.0
)