PEtab.jl

The PEtab.jl package allows for loading models defined in the PEtab standard for parameter estimation into Julia. In addition to datasets of recorded measurements, the PEtab format allows for the specification of a mathematical model for describing said data (which are usually based on Ordinary Differential Equations and provided via SBML files), as well as the final parameter values resulting from the estimation process. Most prominently, the PEtab standard has been used to publish modelling results in Systems Biology so far, see e.g. the Benchmark Collection.

Currently, it is possible to convert a PEtabODEProblem from the PEtab.jl package into a DataModel, (or a ConditionGrid if it consists of more than one condition) via by applying the DataModel constructor. For instance, for the Böhm model with .yaml saved under BöhmYamlPath:

using InformationGeometry, PEtab, Plots
Böhm = PEtabODEProblem(PEtabModel(BöhmYamlPath); gradient_method=:ForwardEquations, hessian_method=:ForwardDiff)
DM = DataModel(Böhm; FixedError=true)

This will automatically extract a simplified representation of the dataset. If error models are used in the PEtabODEProblem to estimate the data uncertainties, they are currently dropped and the uncertainties are fixed to the values dictated by the error model at the best fit values of the error parameters. However, since the likelihood function and its gradient are directly accessed from the given PEtabODEProblem, this does not affect optimisation (such as during profile likelihood computation or multistart optimisation), where changes in the given error parameters are properly accounted for.

Therefore, this mainly affects plots of the datasets from the DM, if further changes to the values of the error parameters have been made after the import of the PEtabODEProblem to a DataModel or ConditionGrid, as these will currently not be visible in said plots of the dataset.

Most other functionality of InformationGeometry.jl should be unaffected by this. In particular, it should be possible to compute ParameterProfiles, MultistartFits and so on without issue.

plot(DM; Confnum=0)
R = MultistartFit(DM; N=5000)
P = ParameterProfiles(DM; N=20, meth=NelderMead())

Note

For PEtabODEProblems consisting of multiple conditions, only the gradient method :ForwardEquations is currently supported. For the Hessian method, the three options :ForwardDiff, :BlockForwardDiff and :GaussNewton are supported.