Explore Our Blog

Automated Population PK Model Selection using Software called ‘DARWIN’

In population PK analyses we build models of drug concentration (pharmacokinetic) and/or drug response (pharmacodynamic) over a time period to understand how different factors influence the body’s interaction with a drug and vice versa. Population PK modeling and clinical trial simulations are an integral part of overall drug development but these analyses are complex. The single most time consuming and expensive part of a population PK analysis is the model selection.

DARWIN – Automated Population PK Model Selection Software

DARWIN is a software program that automates the selection of a population PK model.  DARWIN is based on the computational mathematics of survival of the fittest and evolution known as a ‘Genetic Algorithm’.  For pharmacokinetic drug development model selection, DARWIN automatically looks for continuous improvement in the combination of the pharmacokinetic features (number of compartments, absorption model, covariates, etc.), until ultimately finding the optimal combination of those features and the “best” model.  Combined with high-end computing power, DARWIN can evaluate 15,000 models in less time then it takes to examine 100 models using the traditional approach to manual model selection.

DARWIN is the next generation in population PK modeling with results that are: Robust, Faster, Objective, Reproducible and Cost Effective

How do you use DARWIN for automated Population PK model selection?

  • The process starts, as should all population PK analyses, with human biology. A “search space” is constructed to include all the model features that could be considered, all the hypotheses that could possibly make sense, given what is known about the drug. Examples of items in a “search space” include: is elimination linear or not, what are the possibilities for the absorption mechanism, what demographic characteristics could possibly have an influence on the parameters, etc. All the PK/PD relationships imaginable are assembled into the search space.
  • Next, we get a simple NONMEM model running. This serves two purposes. First, it allows you to check the data set by making basic diagnostic plots. Second, it allows you to get initial estimates for basic PK parameters.
  • Next, we build a list of all the options that might be added to this basic model. All the possible absorption models, all the possible elimination models, all the possible between subject variance terms, and all the possible covariates.
  • We always include different initial estimates for the model parameters in the global model search. This is very useful in avoiding problems with local minima in the NONMEM minimization algorithm. This is an often overlooked aspect of population modeling that is easy to include in a global search approach in a very robust way.
  • Then, we run DARWIN and let it tell us what combination of hypotheses in different aspects of the biology and pharmacology (i.e., these model features) are most consistent with the data.
  • We always go back and reexamine the model and the data, look at lots of plots, and ask more questions about our understanding of the biology and pharmacology. Sometimes we find that no combination of the hypotheses completely explain the data. Then we go back to the basic biology, ask more questions, develop additional hypotheses, and evaluate what else might explain the data. We then use these new hypotheses to refine the search space and re-run DARWIN until an optimal model is identified.

Contact us and learn more about DARWIN and how it can be used efficiently and cost-effectively for your drug program