Visual Predictive Checks (VPC)


knitr::opts_chunk$set(fig.dpi    = 96,
                      fig.align  = 'center', 
                      fig.height = 4, 
                      fig.width  = 6,
                      out.width  = '75%',
                      comment    = '',
                      message    = FALSE)


The Visual predictive checks (VPC) shown below are only intended to demonstrate the various options available in xpose and should not be used as reference for modeling practice. Furthermore, the plots are only based on 20 simulations to minimize the computing time of the examples, the size of the xpdb_ex_pk object and of the xpose package in general.


VPC can be created either by:

  1. Using an xpdb containing a simulation and an estimation problem
  2. Using a PsN generated VPC folder

The VPC functionality in xpose is build around the vpc R package. For more details about the way the vpc package works, please check the documentation website.


The VPC computing and plotting parts have been separated into two distinct functions: vpc_data() and vpc() respectively. This allows to:

xpdb_ex_pk %>% 
 vpc_data() %>%  # Compute the vpc data
 vpc()            # Plot the vpc

The generated VPC data is stored in the xpdb under specials datasets and can be used later on.

xpdb_w_vpc <- vpc_data(xpdb_ex_pk) # Compute and store VPC data

xpdb_w_vpc # The vpc data is now listed under the xpdb "special" data

vpc(xpdb_w_vpc) # Plot the vpc from the stored data

Multiple VPC data can be stored in an xpdb, but only one of each vpc_type.

xpdb_w_multi_vpc <- xpdb_ex_pk %>% 
 vpc_data(vpc_type = 'continuous', opt = vpc_opt(n_bins = 6, lloq = 0.1)) %>% 
 vpc_data(vpc_type = 'censored', opt = vpc_opt(n_bins = 6, lloq = 0.1))

vpc(xpdb_w_multi_vpc, vpc_type = 'continuous')
vpc(xpdb_w_multi_vpc, vpc_type = 'censored')

Common options

Options in vpc_data()

xpdb_ex_pk %>% 
 vpc_data(vpc_type = 'censored', stratify = 'SEX',
          opt = vpc_opt(bins = 'jenks', n_bins = 7, lloq = 0.5)) %>% 

Options in vpc()

Creating VPC using the xpdb data

To create VPC using the xpdb data, at least one simulation and one estimation problem need to present. Hence in the case of NONMEM the run used to generate the xpdb should contain several$PROBLEM. In vpc_data() the problem number can be specified for the observation (obs_problem) and the simulation (sim_problem). By default xpose picks the last one of each to generate the VPC.

# View the xpdb content and data problems

# Generate the vpc
xpdb_ex_pk %>% 
 vpc_data(vpc_type = 'continuous', obs_problem = 1, sim_problem = 2) %>% 

Creating the VPC using a PsN folder

The vpc_data() contains an argument psn_folder which can be used to point to a PsN generated VPC folder. As in most xpose function template_titles keywords can be used to automatize the process e.g. psn_folder = '@dir/@run_vpc' where @dir and @run will be automatically translated to initial (i.e. when the xpdb was generated) run directory and run number 'analysis/models/pk/run001_vpc'.

In this case, the data will be read from the /m1 sub-folder (or if compressed). Note that PsN drops unused columns to reduce the simtab file size. Thus, in order to allow for more flexibility in R, it is recommended to use multiple stratifying variables (-stratify_on=VAR1,VAR2) and the prediction corrected (-predcorr adds the PRED column to the output) options in PsN to avoid having to rerun PsN to add these variables later on. In addition, -dv, -idv, -lloq, -uloq, -predcorr and -stratify_on PsN options are automatically applied to xpose VPC.

The PsN generated binning can also applied to xpose VPC with the vpc_data() option psn_bins = TRUE (disabled by default). However PsN and the vpc package work slightly differently so the results may not be optimal and the output should be evaluated carefully.

xpdb_ex_pk %>% 
 vpc_data(psn_folder = '@dir/run001_vpc', psn_bins = TRUE) %>% 

Try the xpose package in your browser

Any scripts or data that you put into this service are public.

xpose documentation built on June 30, 2021, 9:07 a.m.