How To

Here are some how-to guides. This section should be fairly self-contained and will cover some of the steps of a Bayesian workflow implemented on a nonlinear mixed effects model. You should read the article and stay tuned for the book of the same name.

For this section, I’ll assume a one-compartment model with first-order oral absortion and first-order elimination and some covariate effects - it’s a simple model that everyone knows, so we can concentrate on the workflow and not worry about understanding the model.

  1. Simulating fake data1
  2. Fitting the model
    1. Prior Predictive Checks
    2. Fitting the model
  3. Post-Processing
    1. MCMC diagnostics
    2. Model diagnostics - this includes VPCs and NPDEs.
    3. Posterior Summaries
    4. Comparing the prior and the posterior
  4. Making Predictions
    1. Make predictions for already-observed subjects.
    2. Make predictions for potential future subjects and simulate new subjects/trials.
  5. Cross-Validation
    1. Leave-one-out cross-validation
    2. Leave-one-group-out cross-validation
Back to top

Footnotes

  1. This step can be skipped if you want, but I think it’s an important step to help me understand the model (structural and statistical), especially if it’s a model I’ve never written before, to make sure I’m coding it correctly, and to give me data with a known ground-truth data-generating process and parameters so that I can assess whether the model can possibly work - if it won’t work on fake data that is generated from itself, then it won’t work on real data.↩︎