KeyError and potential solution in sampling tutorial

[adamkova]

Dear Pierre and colleagues,

First of all, thank you for the great tool, the neat support in the docs and the informative articles.

A minor issue/suggestion:
In tutorial_sampling.py, using model iJO1366.json, I got a KeyError on line 72, while applying flux directionality. Going into the code in /analysis/manipulation.py, I found that if I change the end of the code on line 70 from
backward_use.variable.lb = round(solution.x_dict[backward_use.name])
to
backward_use.variable.lb = round(solution.x_dict[backward_use.id])
and similarly for the other three lines as well, I can enforce the use of the raw id name, without the "BU_" and "FU_" prefixes.

Not sure if it was a common issue, but in case anyone else would run into the same, this solved my problem.

Keep up the great work!

Best, Adam

[psalvy]

Good day Adam!

Thank you for the kind words :) We appreciate that people read our prose and use our tools !

Indeed, I forgot to adapt this part of the code to the new fields of the solution attribute. The problem of your solution is that backward_use.id will yield for example PGI instead of BU_PGI, and PGI is the variable name of the reaction PGI -- which might create some issues since that would mean bounding an integer with continuous values ;) (This also explains why you do not get a key error)

However, i replaced solution.x_dict by solution.raw, which yields the raw Series of variable values at solution - this should work. Try pulling from dev (commit b9f14d2) - and tell me if that solves your problem :)

Cheers,
Pierre

[adamkova]

Hi Pierre,

Appreciate the quick response and the update! You are right, I was binding the integers inappropriately... After pulling the update, it works like a charm, thanks.

Subsequently, I found x_dicts to be replaced in /analysis/variability.py as well. Would the below changes make sense (as per your previous update)?
77: profiles[iter_count] = solution.x_dict → profiles[iter_count] = solution.raw
207: fluxes = solution.x_dict[reaction_id] → fluxes = solution.raw[reaction_id]
210: deltag = pd.Series({x.id:solution.x_dict[x.name] → deltag = pd.Series({x.id:solution.raw[x.name]

In the case of line 207, I am a bit confused about the deprecated x_dicts from cobrapy (which were suggested to be changed to fluxes) and these x_dicts that should be changed to raw. Can you maybe suggest an intuitive way to differentiate them? This pytfa - cobrapy integration and tracking the attributes/variables gets a bit tricky for me at certain points... :)

Thanks ahead!
Adam

[psalvy]

Good day Adam,

Thank you for pointing these out!
Yes, I kinda coded this in a hurry without proper documentation ... Here was my reasoning:

• solution.fluxes in cobrapy is a transformed version of the solver output, as it actually calculates the net flux of each reaction by substracting the reverse variable value to the forward variable value. This should be used anytime one needs the actual flux value

• solution.raw is an attribute I added to get a clear copy of the solver output. From there one can access the value at solution for all the variables of the problem. However, looking for a reaction ID in there will only give the forward flux. This should be used for any other variable than fluxes.

I did not want to remove fluxes from solution.raw because then, well, it would not be raw anymore ...
However if you have suggestions with respect to this I would gladly take them !
I should have corrected the last remnants of x_dict in the latest push (dev, commit a985ce8)

Let me know if that solved your plroblem, and I'll close the issue :)

Cheers,
Pierre

[adamkova]

Good day to you too, Pierre,

Thank you for the new update and the detailed explanation. The pace at which you addressed these issues is impressive 👍.

This thumb rule about the use cases of fluxes and raw makes good sense to me now. I also agree that for completeness, solution.raw should contain everything.

The changes seem to have done the trick, the code regained its flow.

Regarding suggestions, I would not make any now. It is a big project, and I could see that it takes quite some time to get a good enough overview of what is happening under the hood. Momentarily, it seems more feasible to assume that you made all the right choices during coding. :)

With that, please feel free to close this issue.

Keep it up,
Adam