overflow

Proteome-constrained enzyme-constrained model analysis of overflow metabolism in Saccharomyces cerevisiae, across four carbon/nitrogen ratios and one high growth rate condition.

The analysis is built on the GECKO 4 ecYeastGEM and runs in Python through geckopy and raven-toolbox. The MATLAB implementation that produced the published results, which used GECKO 2, is kept in legacy_matlab/.

Installation

pip install -e ".[dev]"

Python 3.11 or later. Gurobi (gurobipy with a licence) is the default solver for the analysis and the tests. Another solver can be chosen with --solver, or with the OVERFLOW_SOLVER environment variable, which also applies to the tests; an unavailable solver is an error, not a silent switch.

Layout

Path	Contents
data/	Measured proteomics, fermentation rates, ribosome subunits, annotation
models/	The GECKO 4 ecModel and the conventional GEM it derives from, see models/PROVENANCE.md
src/overflow/	The analysis package
tests/	Test suite
results/	Output tables and figures
legacy_matlab/	The GECKO 2 MATLAB implementation and its results

Experimental data

data/fermentationData.txt holds, per condition, the total protein content, the dilution rate and the measured exchange rates. data/abs_proteomics.txt holds absolute protein abundances in mmol/gDW: three biological replicates per condition, four for hGR.

from overflow import load_conditions

conditions = load_conditions()
conditions["CN4"].d_rate          # 0.1
conditions["CN4"].byproduct_bounds()   # undetected byproducts are blocked

Model

from overflow import build_adapter, load_model

model = load_model(build_adapter(conditions["CN4"]))

build_adapter reads model_adapter.toml and layers the condition's measured protein content and dilution rate on top.

Building the condition models

python -m overflow.build                 # every condition
python -m overflow.build CN4 --solver glpk   # instead of the default, Gurobi

Each condition gets models/ecModel_P_<cond>.yml and, in results/, the enzymes whose measured abundance the model could not run on, the enzymes whose caps had to be released, and the full flux distribution.

Options change what the models are asked to account for:

• --fit-rates holds CO2, oxygen and the byproducts within a tolerance of their measurements (--rate-tolerance, default 5%) and raises the measured abundances by the least that makes that feasible. Without it those rates are free, and the model disposes of surplus carbon through whichever exit is cheapest in protein rather than respiring it.
• --uptake-flex caps glucose uptake at that multiple of the measured rate (default 1.05). CN4 needs 1.08: at its dilution rate the model needs 6.5% more glucose than was measured, even with protein unlimited.
• --objective chooses what the finished model optimises: the smallest protein pool (protein, the default), or the smallest total flux at the dilution rate over every reaction (flux) or over the metabolic reactions only (metabolic-flux). The flux objectives need --fit-rates.
• --block RXN ... closes the named reactions before anything else is applied. The analysis blocks r_2129, the proton leak: it has no enzyme and no bound, so it is a free sink for oxygen. Fitting the measured oxygen with it open sent most of the respiration through it at high C/N (P/O 0.19 at CN75, against 1.1-1.2 with it closed), because the fit raises each cap only as far as the rates need and the built models can carry no maintenance ATP. Closed, the respiration is coupled and the maintenance shows up as non-growth ATP use.
• --scale-protein rescales biomass to the measured protein content. Off by default: the model has no carbon to spare at the measured glucose uptake, and CN4 then falls short of its dilution rate.

Adding the ribosome

python -m overflow.build_ribosome

Writes models/ecModel_P_<cond>_ribosome.yml. The protein pseudoreaction is split so that protein comes out of a translation reaction catalysed by the ribosomal subunits, at 10.5 amino acids per second per ribosome. The core is the 48 subunits whose average abundance across all conditions reaches 1e-5 mmol/gDW.

Summarising enzyme usage

python -m overflow.analyze_usage

Writes, per enzyme and condition, how much of it the model uses and what fraction of what was available that is, plus the capacity usage of the annotated systems and the two figures over them.

Random sampling

python -m overflow.run_sampling --procs 16

Samples the conventional model under the measured rates, with and without formate in the measured set, and writes the sampled means and standard deviations, the byproducts the model secretes when it is not told to, and the ATP and redox budget.

Sampling draws vertices by maximising small random objectives and then minimising total flux at each, as the published analysis did; means over such draws are vertex means rather than an average over the interior of the flux space. --no-min-flux skips the second step, which lets the sampler wander into high-flux routes: the pentose phosphate pathway then carries several times the published flux.

Conditions are independent, so they can be sampled as separate runs (for example one Slurm job each) and combined afterwards:

python -m overflow.run_sampling CN4 --procs 12 --results-dir runs/CN4   # one per condition
python -m overflow.run_sampling --merge runs/CN4 runs/CN22 runs/CN38 runs/CN75 runs/hGR

The loop-free screening that precedes sampling has a five-minute limit and retries with a new seed, because cobra's search for cyclic reactions occasionally runs for hours.

Comparing against the published results

python -m overflow.compare

Writes results/COMPARISON.md: predicted exchange rates, median capacity usage per system and the sampled ATP and redox budget, this pipeline beside the MATLAB one, read from legacy_matlab/results/.

Running the whole analysis

The analysis is run with --fit-rates, implemented in overflow.build (relax_to_measured_rates): it holds CO2, oxygen and the byproducts at their measurements and raises the measured enzyme abundances by the least that makes that feasible. Without it those rates are free, and on the earlier tutorial model the enzyme-constrained solution disposed of surplus carbon through unmeasured exits instead of respiring it, so the flux distribution did not describe the measured physiology. --uptake-flex 1.08 gives CN4 the glucose it needs, and --block r_2129 closes the proton leak (see the build options).

python -m overflow.build --fit-rates --rate-tolerance 0.08 --uptake-flex 1.08 --block r_2129
python -m overflow.build_ribosome
python -m overflow.analyze_usage
python -m overflow.run_sampling --procs 12
python -m overflow.compare

Each step reads the models the previous one wrote, so --models-dir and --results-dir keep a run self-contained.

Tests

pytest                       # everything except the full condition builds
pytest -m "not slow"         # skip the tests that load a genome-scale model
pytest -m integration        # build a condition end to end; minutes

The suite runs on the same solver as the analysis, Gurobi by default. Set OVERFLOW_SOLVER=glpk to run it without a licence, as CI does.