JuPOETs: a constrained multiobjective optimization approach to estimate biochemical model ensembles in the Julia...

Area of Science:

• Computational Biology
• Mathematical Modeling
• Software Development

Background:

• Ensemble modeling offers robust predictions by utilizing parameter or model families to manage uncertainty in deterministic mathematical models.
• This approach contrasts with single best-fit parameters or fixed model structures, enabling better confidence intervals and constrained predictions.
• Parameter ensembles can be selected based on simulation error, diversity, and steady-state performance.

Purpose of the Study:

• Introduce the Pareto Optimal Ensemble Technique (JuPOETs) in Julia for estimating parameter or model ensembles.
• Demonstrate JuPOETs' capability in handling multiobjective optimization problems and biochemical model identification.
• Highlight JuPOETs' efficiency and performance compared to existing implementations.

Main Methods:

• Integrated simulated annealing with Pareto optimality to estimate ensembles on or near the optimal tradeoff surface.
• Applied JuPOETs to a suite of multiobjective test functions with parameter bounds and system constraints.
• Utilized JuPOETs for the identification of a proof-of-concept biochemical model with four conflicting training objectives.

Main Results:

• JuPOETs identified optimal or near-optimal solutions approximately six-fold faster than a comparable Octave implementation for test functions.
• For a biochemical model, JuPOETs generated parameter ensembles that accurately represented conflicting datasets.
• The technique simultaneously estimated parameter sets performing well on individual, conflicting objective functions.

Conclusions:

• JuPOETs is a powerful tool for estimating parameter and model ensembles via multiobjective optimization.
• The algorithm is adaptable to various problem types, including mixed-variable, bilevel, and constrained optimization.
• JuPOETs is open-source, available under an MIT license, and installable via the Julia package manager.