This GitHub repository contains scripts for running PEtab parameter estimation benchmark using PyPesto (via the AMICI interface) and the Julia PEtab importer (the package does not have a cool name yet).
Given i) a Julia 1.8.5 executable (or later version) at path_to_julia, ii) and the AMICI dependencies PyPesto and PEtab.jl can be installed by running bash setup.sh from the project root directory.
- Note : the Julia executable path must be set manually in setup.sh.
- Note : the compilation time for Julia can be significant.
Given that the setup was successful a benchmark can be run by executing bash Run_benchmark.sh modelName, so for example bash Run_benchmark.sh Boehm_JProteomeRes2014 will run the benchmark for the Boehm model.
1 Note : In the Master_thesis/Benchmark/Run_benchmark.sh file the Julia executable path must be set manually.
The available optimization algorithms that can be provided for the Julia are:
| Optimizer | Description |
|---|---|
| IpoptAutoHess | Ipopt using full hessian (via autodiff) |
| IpoptBlockAutoDiff | Ipopt using block approximated hessian (via autodiff) |
| IpoptLBFGS | Ipopt using L-BFGS hessian approximation |
| OptimIPNewtonAutoHess | Optim.jl interior point Newton full hessian (via autodiff) |
| OptimIPNewtonBlockAutoDiff | Optim.jl interior point Newton via bloack approximated hessian (via autodiff) |
| OptimIPNewtonGN | Optim.jl interior point Newton with Gauss-Newton hessian approxmiation |
| OptimLBFGS | Optim.jl L-BFGS method |
| FidesAutoHess | Fides using full hessian (via autodiff) |
| FidesBlockAutoHess | Fides using block approximated hessian (via autodiff) |
| FidesGN | Fides using Gauss Newton hessian approximation. |