ExaModelsPEtab.jl

Formulates ExaModels from PEtab parameter-estimation models.

Unlike PEtab.jl, which uses the sequential method for dynamic optimization, ExaModelsPEtab.jl applies the simultaneous method (orthogonal collocation) to formulate the problem as an NLP. The ExaModel can then be solved with MadNLP using a CPU or GPU backend.

Usage

An ExaModel is formulated in a single function call:

examodel = petab_examodel(
  filename::String;
  backend = nothing,
  K::Int = 4
)

Keyword arguments

• backend — Array backend: nothing for CPU (default) or CUDA.CUDABackend() for GPU.
• K — Degree of the Lagrange interpolating polynomial (number of collocation points per interval); defaults to 4. Steady-state models ignore K since no mesh is required to solve f(z, p) = 0.

Example

using ExaModelsPEtab, MadNLP

# Build the ExaModels NLP from a PEtab problem YAML file.
model_CPU = petab_examodel("path/to/problem.yaml")   # CPU
result_CPU = madnlp(model_CPU)

# Build the ExaModel using CUDA backend and solve with GPU solver
using CUDA, MadNLPGPU
model_GPU = petab_examodel(
  "path/to/problem.yaml";
  backend = CUDA.CUDABackend(),
  K = 4
)
result_GPU = madnlp(model_GPU; tol = 1e-6)

References

1. Shin, S., Anitescu, M., & Pacaud, F. (2024). Accelerating optimal power flow with GPUs: SIMD abstraction of nonlinear programs and condensed-space interior-point methods. Electric Power Systems Research, 236, 110651.

2. Persson, S., Fröhlich, F., Grein, S., Loman, T., Ognissanti, D., Hasselgren, V., Hasenauer, J., & Cvijovic, M. (2025). PEtab.jl: advancing the efficiency and utility of dynamic modelling. Bioinformatics, 41(9), btaf497.

3. Schmiester, L., Schälte, Y., Bergmann, F. T., Camba, T., Dudkin, E., Egert, J., Fröhlich, F., Fuhrmann, L., Hauber, A. L., Kemmer, S., et al. (2021). PEtab—Interoperable specification of parameter estimation problems in systems biology. PLoS Computational Biology, 17(1), e1008646.