Tutorials

The PEtab.jl tutorials cover how to set up parameter-estimation problems with various features, and how to run and configure parameter estimation.

Creating a mechanistic parameter estimation problem

• Simulation conditions: Measurements collected under different experimental conditions (e.g. simulations use different initial values).

• Pre-equilibration: Enforce a steady state before the model is matched against data (pre-equilibration).

• Simulation condition-specific parameters: Subset of model parameters which are estimated take different across simulation conditions.

• Observable and noise parameters: Observable/noise parameters in PEtabObservable formulas that are not part of the model system (e.g. scale/offset), optionally time-point-specific.

• Events/callbacks: Time- or state-triggered events/callbacks.

• Import PEtab standard format: Load problems from PEtab standard format.

• Supported model systems: Supported ways to define the dynamic model (ReactionSystem, ODESystem, or ODEProblem) and available options.

Parameter estimation for mechanistic models

• Parameter estimation extended tutorial: Extended tutorial on estimation functionality (e.g. multi-start, Optimization.jl integration).

• Available optimization algorithms: Supported and recommended algorithms.

• Plotting parameter estimation results: Plotting options for parameter estimation output.

• Model selection: Automatic model selection with PEtab-Select.

• Wrapping optimization packages: How to use a PEtabODEProblem directly with an optimization package such as Optim.jl.

• Bayesian inference: Sampling-based inference (e.g. NUTS and AdaptiveMCMC).

Creating a SciML parameter estimation problem

• SciML starter tutorial: Introductory tutorial on how to create and train a SciML problem that combines mechanistic ODE and ML (neural network) models.

• ML models in observables: define an ML model in the observable formula of a PEtabObservable (e.g. to correct model misspecification).

• Pre-simulation ML models: define ML models that map input data (e.g. high-dimensional images) to ODE parameters or initial conditions prior to model simulation.

• Importing PEtab SciML: import problems in the PEtab-SciML standard format.

• Training strategies: How to improve SciML training (parameter estimation) performance using strategies such as curriculum learning and multiple shooting via PEtabTraining.jl.