Setting medium to itself produces different results

[joeryjoery]

Problem description

I was changing around some of the variables within the model.medium; I found that setting the medium using model.set alters the optimization objective.

By setting the medium of model to itself, which shouldn't change anything, the output of the optimization changes.

Edit: added cobra import and model name.

Model: BiGG Model iNF517 - http://bigg.ucsd.edu/models/iNF517

Code Sample

Create a minimal, complete, verifiable example.

import cobra

model = cobra.io.read_sbml_model("iNF517.xml")

with model:
    print(model.slim_optimize())

with model:
    model.medium = model.medium
    print(model.slim_optimize())

Context

``` 0.04263460544337327 0.0746105595259032 ```

[Midnighter]

Hi @joeryjoery,

You'll need to be a little bit more specific. What change do you observe exactly? In which model is this? Changes within the solver's tolerance, for example, can be expected.

[joeryjoery]

Hi,

I'm using the iNF517 model: http://bigg.ucsd.edu/models/iNF517. (I edited it into the post).

What I expect is happening is that the default model medium has asymmetric bounds within the exchange reactions. I see in the source @medium.setter (line 253: https://github.com/opencobra/cobrapy/blob/devel/src/cobra/core/model.py) that the bounds are overriden.

I am not aware of what exactly is being overriden behind the scenes

I would expect that this isn't intended behaviour...

[cdiener]

I can reproduce that.

In [39]: import cobra
    ...: 
    ...: model = cobra.io.read_sbml_model("iNF517.xml")
    ...: 
    ...: with model:
    ...:     print(model.optimize())
    ...: 
    ...: with model:
    ...:     model.medium = model.medium
    ...:     print(model.optimize())
    ...: 
<Solution 0.043 at 0x7fb3a251a940>
<Solution 0.075 at 0x7fb3a2488910>

The issue is that this model has fixed export bounds. In particular:

EX_ac_e        (0.83, 1.92)
EX_etoh_e      (0.44, 1.02)
EX_for_e       (1.71, 3.94)
EX_lac__L_e     (0.4, 0.92)
EX_pro__L_e    (0.03, 0.44)

which will be converted to

EX_ac_e        (0, 1.92)
EX_etoh_e      (0, 1.02)
EX_for_e       (0, 3.94)
EX_lac__L_e    (0, 0.92)
EX_pro__L_e    (0, 0.44)
dtype: object

The problematic behavior is how the medium setter manages exchanges not in the medium. Right not those get set to (0, default_bound) which overwrites the export bounds. I can send a fix.

Though for that particular case this is probably kind of a bad set of defaults. Fixing secretion rates independent of the actual growth medium is likely not going to give you good results. It's kind of a bad decision by BIGG to keep those in the model. In this case, it cuts the maximum growth rate almost by half.