Code and data for "Multiobjective design of growth media with genome-scale metabolic models and Bayesian optimization"

Code and data for “Multiobjective design of growth media with genome-scale metabolic models and Bayesian optimization” by Hallmann et al. Includes the full implementation of the multi-objective Bayesian optimization framework, processed simulation data, analysis scripts, and figure-generation notebooks required for reproducibility of the study. The ZIP file contains: The full optimization pipeline (simulation and gsMOBO loop), Processed data used to generate the figures in the associated manuscript, Helper notebooks for adapting the method to your own model. File structure Top-level layout: mobo-medium-conditions/ ├── analysis/ ├── data/ ├── optimization/ ├── .DS_Store ├── .gitignore └── README.md analysis/ Jupyter notebooks and helper code for analyzing results and generating figures. Contains: Figure-generation notebooks: Notebooks that read pre-processed .csv data (from data/) and generate the figures appearing in the main text and supplementary material. Post-processing notebooks / scripts: Code that: Reads the complete simulation data (produced by the optimization workflow), Extracts the relevant subsets, Writes compact .csv files into data/ for plotting. data/ Processed data used to build figures. Contains: CSV files (*.csv): Pre-processed subsets of the full simulation results (e.g., selected conditions, Pareto fronts, summary metrics), Data structured in a way that is convenient for the plotting notebooks in analysis/. optimization/ Full code needed to run the Bayesian optimization and simulations. Main pipeline notebook (referred to in the original README as Main_MOBO.ipynb): Loads your metabolic / mechanistic model (via CobraPy or compatible tools), Defines medium components, bounds, and costs, Configures and calls the gsMOBO routine (e.g. media_BayesOpt), Runs simulations and stores outputs. Auxiliary notebooks / scripts: Utility functions for: Building design spaces, Evaluating objectives (growth, production, cost), Handling constraints and batch evaluations, Helper routines for saving simulation outputs in a structured format consumed by analysis/. Set up a Python environment Create a virtual environment (via venv or conda) and install the dependencies used in analysis/requirements.txt Adapting the method to your own project The repository is designed so that you mostly only need to edit Main_MOBO.ipynb to apply the approach to a new system. Step 1 — Load your model In optimization/Main_MOBO.ipynb: Load or construct your metabolic / process model using CobraPy or a compatible library. If necessary, adapt your model so that: Reaction and metabolite naming follows CobraPy conventions, Key fluxes (biomass, product) are clearly defined. Step 2 — Define the design space (medium components) Still in Main_MOBO.ipynb: Specify the medium components you want to optimize (e.g., carbon sources, nitrogen sources, trace elements). For each component, define: Bounds (minimum/maximum concentrations or fluxes), Cost (per unit), if medium cost is one of your objectives. If a cost is not provided, default values are used (as per the original project description). ([GitHub][1]) Step 3 — Configure objectives Choose the combination of objectives you care about, e.g.: Growth (maximize biomass flux / growth rate), Production (maximize flux of a particular reaction representing your product), Cost (minimize total medium cost). You can include any subset or combination, as long as they are compatible with the model and optimization routine. Step 4 — Run the Bayesian optimization (media_BayesOpt) From Main_MOBO.ipynb: Call the main routine (referred to in the original README as media_BayesOpt) with: The model, Design space definition (medium components, bounds, costs), Objective configuration and weights (if needed), BO hyperparameters (number of iterations, batch size, kernel settings, etc., depending on implementation). The function will: Propose medium compositions to test, Evaluate objectives through simulations (or experiments, if connected), Update the surrogate model and acquisition function, Iterate until the stopping criterion is met. Step 5 — Export and analyze results Save the optimization traces and final Pareto fronts. Use or extend the analysis/ notebooks to: Inspect trade-offs between objectives, Visualize medium compositions and model behavior, Export .csv datasets to data/ and regenerate figures in figs/.