DifferentiableMetabolismCode

The code in this repo can be used to reproduce all of the results in the associated paper Interrogating the effect of enzyme kinetics on metabolism using differentiable constraint-based models (preprint here).

There are three main results in the paper:

1. Sensitivity analysis of an enzyme constrained GECKO model
2. Gradient descent on measurements to fit enzyme turnover numbers
3. Sensitivity analysis on a model that incorporates full Michaelis-Menten kinetics

Note, the Julia environment used for the analyses can be recreated by activating the Project.toml file, as described here. The directory data contains all the data (turnover numbers, Michaelis constants, experimental data, etc.), as well as the scripts used to parse them into the format used in this work. The directory model_construction contains scripts used to download the metabolic models, as well as add stoichiometric data for enzyme subunits, enzyme turnover numbers, and enzyme masses to these models (called fixed_models in the context of this work).

If anything is unclear, please file an issue on this repo.

1. Sensitivity analysis of a GECKO model

The directory analyses/gecko contains all the scripts used to generate the associated results. In particular, the script gecko_iml1515.jl can be used to run the sensitivity analysis of the flux and concentration predictions to enzyme turnover numbers. Note, it will attempt to save the images created in the script to another folder not contained in this repo (the repo used to write the paper), so comment those image file saving lines out.

2. Gradient descent to fit turnover numbers

The directory analyses/gd_gecko contains all the scripts used to generate the associated results. The script cluster.jl can be used to run the gradient descent algorithm for a specific master_id, which is the ID of the respective experimental data. Valid IDs are taken from the data source (see the paper), but include WT1#B1, WT2#B2etc.

3. Sensitivity analysis of a Michaelis-Menten constrained model

The directory analyses/crispr contains all the scripts used to generate the associated results. In particular, the file michaelis_menten.jl can be used to generate the results, and plot_mm_all.jl will plot all the sensitivities. In contrast, plot_mm_few.jl can be used to plot a representative selection of the results (which is easier to inspect).