How to create a virtual population from a computational model ?

  • updated 4 wk ago

Once a Computational Model (CM) is imported in jinkō, you can create from it as many virtual populations as needed in order to conduct simulated trials. In jinkō, this process is simplified via the use of an automatically generated framework, called a Vpop Design, comprised of marginal distributions of descriptors issued from the model and of correlations between these descriptors.

Step 1: Create a framework (a 'Vpop Design') and review descriptors' theoretical distributions

From the 'create new' menu, you have the option to create a Vpop Design from a Computational Model, in a few easy steps:

  • Select the Computational Model that will be used to generate the Vpop Design
  • Name your Vpop Design, (optionally) add a description and place it into the working folder of your choice
  • Validate to create your Vpop Design
  • (Please note that you also have at this point the option to directly upload a virtual population of patients, via a .json or a .csv file, should you have ready-data you want to upload to the platform)

By default, the descriptors with an "Patient descriptor" input tag (see 'descriptors input tags' infobox below) will be preselected. For each selected descriptor, you have:

  • their types and input tag (from the CM, uneditable)
  • the default value and unit (from the CM, uneditable)
  • the chosen distribution law and associated parameters, where a default is generated from the default value (editable and specific to the Vpop Design). The theoretical distribution is plotted on the right-hand side, to verify that it corresponds to your expectations (you can also download these plots in a .png format)
  • a reference, typically a link to track where the values come from, if relevant (editable and specific to the Vpop Design)

create vpop

Available distributions:
  • Uniform
  • Normal
  • NormalTruncated
  • LogUniform
  • LogNormal
  • LogNormalTruncated
  • Mixture
  • Bernoulli
  • Binomial
  • Categorical
  • Poisson
  • SkewNormal
  • Weibull
Descriptors input tags

Descriptors input tags, appearing in green under the title of each baseline descriptor, are used to categorize its type of model input based on its nature, there are 7 different types:

  • Model Intrinsic: when the baseline descriptor is assumed to have an unique value independent of the individual
  • Patient Descriptor Known: when the baseline descriptor is assumed to have a value specific to each individual and whose values are known for each of them
  • Patient Descriptor Partially Known: when the baseline descriptor is assumed to have a value specific to each individual and whose value are not known for each of them but whose distribution is known
  • Patient Descriptor Unknown: when the baseline descriptor is assumed to have a value specific to each individual and whose distribution is not known
  • Formulaic: when the baseline descriptor is a formula whose value depends on the value of other parameter(s)
  • Protocol Specific: when the baseline descriptor does not possess an intrinsic value for each individual but whose value depends on external factors that may be changed in scenarios
  • Technical: when the baseline descriptor is not a biological parameter and does not fit into one of the above descriptions

Input tags cannot be edited in a Vpop Design.

Step 2: Edit Correlations

Before creating a virtual population of patients from your Vpop Design, you also have the possibility to edit correlations between several descriptors for which you have defined a distribution. Correlations are statistical rules on the relations between two descriptors based on what is generally observable in a population (from the literature, from internal knowledge...). For instance, a positive correlation between the sex type (male in this example) and a smoking status (smoker / non-smoker) would indicate that in the population generated we would observe more male than female smoking (virtual) patients. You can create as many correlations pairs as you feel relevant for your design.

create vpop

Step 3: Create a Virtual Population ('Vpop')

Once you feel that you have the right virtual population design, you can start creating your virtual patients cohorts (each cohort is called a Vpop in jinkō). You can create as many Vpops as you want to explore and run trials with different variations of populations.

For each new Vpop, you need to chose a name and optional description, and a size (number of patients) and a seed. The seed is typically used for reproducibility of data: the sames Vpop design with the same size and the same seed will give exactly the same Vpop, and changing the seed allows to re-sample a population using the same codistribution. You can also select the Variance reduction option, which allows to reduce by half the number of samples drawn by using the antithetic path. 

Once you have created a Vpop, you can visualize its distribution over the theoretical distributions of your design, in order to verify that the distributions used in order to create your vpop correctly follow the intended design. If not, you may want to use a larger size. 


Congratulations! You can now use your virtual population to simulate a trial!

Reply Oldest first
  • Oldest first
  • Newest first
  • Active threads
  • Popular