yannikschaelte / dfba_tests

Parameter estimation and uncertainty quantification for dFBA models

• using dynamic-fba for simulating dFBA models
• using pyPESTO for parameter estimation & UQ

---

Running parameter estimation:

With the function defined in

./examples/ex1_pypesto_bonna.py

you can run an optimization for finding optimal parameters on 3 implemented dfba examples (example1, example1_aerobic, example6 -> chosen from dynamic-fba-repository collection)

The function run_optimization includes:

• Build DFBA model, where the FBA part is defined in SBML in the file "xxx.xml.gz" (model_dir) and the dynamic part is defined in the function 'get_dfba_model_ex1_ex6.py'
• builds a pickable DFBA model (to enable parallelization with pypesto)
• reads data
• defines pypesto-objective
• defines pypesto-problem
• runs optimization
  • optimizers: SLSQP, TNC, L-BFGS-B, Pyswarm, Fides
  • cost-functions:
    • Least Squares
    • Neg-log-likelihood with normal noise model
    • Neg-log-likelihood with laplace noise model
• stores optimization results

Objective function

The objective function is located in: ./pypesto_dfba/optimize_dfba/objective_dfba.py

Plotting results

With the script in examples/plot_results.py you can plot optimization results with a given hdf5-results object. (or a history.hdf5-object, or just an xhat parameter vector)

• plot & save waterfall plot
• plot & save simulation trajectories with data points
• plot & save parameter plots