Remove the boundary compartment from Human-GEM

[JonathanRob]

Description of the issue:

The default Human-GEM model currently contains a boundary compartment [x], used to close mass balances on exchange reactions. For example, the exchange of oxygen between the extracellular compartment [s] and the boundary is formulated as:

O2[s] <=> O2[x]

However it is generally more common among the modeling community to formulate exchange reactions without the balancing boundary metabolite:

O2[s] <==>

I propose that we change the "default" state of the model such that the boundary compartment is removed. Users can then choose to add it back when needed (e.g., for checkTasks or running tINIT) using the addBoundaryMets function:

model = addBoundaryMets(model);

The reason for my suggestions is that the boundary compartment is not really standard practice in the modeling community, and is often a source of confusion and compatibility problems. For example, if FBA is run on the model with boundary metabolites still present, flux through the biomass reaction (biomass_human) will be zero because the model is essentially in "closed" form. Users therefore need to go through the extra step of removing the boundary metabolites before running FBA and many other common analysis methods.

I hereby confirm that I have:

• Checked that a similar issue does not exist already

[haowang-bioinfo]

@JonathanRob sounds good. How about start with a to-do list regarding this change?

[JonathanRob]

It's actually pretty straightforward; just need to run simplifyModel and remove the empty compartment. I've implemented it on the removeBoundaryComp branch, but I will wait to make a pull request until we've taken a bit of time to consider other potential necessary changes/effects, and to complete existing PRs.

[haowang-bioinfo]

I wonder if this removal affects any model manipulation and simulation functions from RAVEN?

[JonathanRob]

I'm only aware of two affected functions that rely on the presence of boundary metabolites (an unconstrained field): checkTasks and getINITModel. The checkTasks function gives a warning if its missing, and the getINITModel function gives an error.

Maybe @edkerk can confirm, but as far as I am aware, RAVEN is designed to work without boundary metabolites.

[edkerk]

RAVEN simulations indeed assume that no boundary compartment is present. @simas232 I remember there was an exercise at a course that relied on the boundary compartment being included, do you remember which RAVEN function was not working without boundary?

Regardless, we can also write a small function that adds a boundary compartment, takes the output of getExchangeRxns and add the corresponding metabolite in the new boundary compartment. Will be super easy to code, as metabolite ID for the [b] metabolite we could take the metabolite ID that is already present in the exchange reaction and append _b?

[JonathanRob]

Ok, that's what I thought, thanks @edkerk.

We actually have a function in this Human-GEM repo that adds boundary metabolites: addBoundaryMets. One option is to move this function into RAVEN instead.

[simas232]

@JonathanRob, @edkerk, RAVEN already has a function which adds a boundary compartment (i.e. closes the model), called closeModel.

@edkerk, I think that only tINIT related functionalities, including fitTasks require a boundary compartment. So we may simply enable closeModel automatically, if an input GEM does not have a boundary compartment.

[edkerk]

@simas232 Excellent, closeModel does exactly what I had in mind. So the changes @simas232 suggested can be applied to RAVEN, and then there is no problem whatsoever to remove boundary mets from Human-GEM.

[JonathanRob]

Ah ok, I didn't even know about closeModel, nice.

@edkerk, I'm not sure it's necessary to make any changes to RAVEN to accommodate the change. Or maybe simply update the existing error/warning messages that are thrown when the unconstrained field is missing so that they mention the closeModel function.

[JonathanRob]

Addressed in #179.