It is almost always possible to build a model to describe a process, but the rates at which interactions occur, and other associated model parameters, are often unknown and undocumented in literature.  Under these circumstances, parameter estimation, also known as model fitting, becomes necessary.

Immunetrics owns a suite of in-house optimization algorithms that allow us to analyze any ordinary differential equation based-model. Families of algorithms include:

  • Custom genetic algorithms
  • Hillclimbing algorithms, such as stochastic, gradient-based, and simplex-based

 

Using these tools, we can fit a model to data and experimental setups, providing estimates for unknown parameters. Qualitative and quantitative heuristics are incorporated as criteria for a successful fit.  For under-constrained systems, results take the form of an ensemble of potential parameter vectors, providing insight into model variability, sensitivity, and robustness.

Using this approach, we can:

  • Fit to ranges of data and/or provide ranges of viable parameter values
  • Fit to multiple experimental setups simultaneously
  • Identify parameter bounds, or insensitive unconstrained parameters
  • Identify key physiological differences between experimental results

To estimate model parameters, we use experimental, clinical trial, and/or critical care data to train our models to reproduce observed results, determine values for model parameters, and enable our models to predict future results beyond the scope of the data.