MBDE (Model-based design of experiments)

A free GitHub account is required to access the program code in this repository; send us your GitHub username.

Module containing some experimental :) calculations to support design of experiments using mechanistic models. If you make improvements, please share them here. The repository is not yet public, so the code should not be shared outside this repository and your own dynamic models.

Experimental only in the sense that we are evaluating how useful the calculated metrics can be for the design of experiments and whether these metrics suggest experiments that are comparable to or even superior in value to those designed by an experienced practitioner.

Experiments with potential high information content may be identified by simulating a list of potential runs (scenarios). Potential experiments may be ranked according to several measures including:

• estimability analysis based on gradients with respect to parameter values evaluated at each time a potential measurement sample is taken
• the diagonal terms of the Fisher Information Matrix, also based on these gradients and the covariance matrix of the fitted parameters
• the determinant, trace and maximum eigenvalue of the Fisher Information Matrix, so-called ‘D’, ‘A’ and ‘E’ optimal designs.

Instructions for use:

• Run a fit in the Fitting window and generate a Fitting Report that indicates the parameter estimates and fitting statistics. Include this in your Excel file as ‘Fitting Report’.
• Insert the code in a VBA Module inside the target model that contains the scenarios and data.
• Insert a reference in your VBA project to ModelAutomation.
• Use your existing scenarios or generate a new list of scenarios (e.g. using a Full Factorial run, or your favourite DOE software) and place all intended scenarios on your model scenarios tab.
• Add a new worksheet called ‘MBDE’ to store the output.

• Then invoke the code from your own code or by pressing a macro button. The main module to invoke is:

  • RunMBDE - generates results on the MBDE worksheet.

Results are in the form of a large table with summary metrics at the top and results at each time point below; confidence bands and gradients at each point are included. Results may be plotted in various ways, including time profiles with confidence bands as error bars.

Additional background information and an example file (send us your GitHub user ID if you don’t already have access to these):

• More details of the calculation methods
• Example model as used in the PDF