feat: ModelAdapter pattern for organism specific parameters and customizations

[johan-gson]

Main improvements in this PR:

Added an adapter pattern that can enable us to use GECKO for different species without copying files. Right now the adapter is pretty small, but the idea is to add more functions in the base class when needed. The idea is then that an adapter is sent into GECKO when generating models, and that GECKO can ask the adapter for information that is accessed differently for each species/model. Examples include parameters, pathways to data, and which reactions are spontaneous. The code is not plugged in anywhere, so this is totally safe.

I also added an installation script, let's see what you think about that. To install (i.e., setup path):
cd to GECKO root
GECKOInstaller.install
to uninstall (i.e., remove GECKO from path):
GECKOInstaller.uninstall

This is practical if you have several versions of GECKO on your disk and want to switch between them.

I hereby confirm that I have:

• Tested my code with all requirements for running GECKO
• Selected devel as a target branch (top left drop-down menu)

[edkerk]

• GECKOInstaller is nice! This should be persistent, right? So after a restart, GECKO should still be in the path. Even people that only have one GECKO instance can then easily add it to the path.
• The adapter solution is very nice as it seems to be very fast as well. I'm not sure how you would now get access to the functions that it also defines?

[mihai-sysbio]

Very nice demonstration of how the adapters are supposed to work. I think @IVANDOMENZAIN might have an opinion here if these adapters should go in the GECKOrepository or in ecModels. Also, I'm not sure why getHumanGEMRootPath needs to be defined outside the adapter.

[johan-gson]

So, to explain how you access the functions in the adapter:
function model = applyKcatConstraints(model,kcatSource,kcatComplex,kcatSubs,updateRxns,modelAdapter)
%...
params = modelAdapter.getSpontaneousRxns(model);
%...
end

Example of how to call this:
adapter = Yeast8Adapter();
x = applyKcatConstraints(..., adapter);

or just
x = applyKcatConstraints(..., Yeast8Adapter());

If you want to change some params in the adapter (which is why I left the params public, which is a bit ugly, but practical):
adapter = Yeast8Adapter();
adapter.params.whatever = 3;
x = applyKcatConstraints(..., adapter);

GECKO here knows nothing about the specialized adapters, the modelAdapter sent in is just an object of a class that inherits the ModelAdapter base class. Let's look at Yeast8Adapter as an example - this class inherits modelAdapter. Since getSpontaneousRxns is abstract, YeastAdapter must implement that function - otherwise you will get an error when you try to create an object of that class. When the call is made, matlab will always use the function in the specialized class first, and GECKO doesn't have to know anything about that implementation in the specialized class - this could for example be placed in another repo. So, in java or c++, this is how you must solve it since those are typed languages. In matlab, you would technically not need a base class to accomplish this, as long as the object sent in has the right functions it will work anyway. But I think it is still better to use a base class - this makes it clear what needs to be implemented in the adapter and is what programmers would expect, and it also allows for default implementations in the base class (see getParameters for example).

[johan-gson]

GeckoInstaller is persistent - it calls savepath, so if you restart matlab the path is still changed. The same could be added to RAVEN, but I think you have something there already perhaps? I always have these installer classes on my disk for Human-GEM, RAVEN and Yeast-GEM

[johan-gson]

About where to place the specialized adapters: I'm thinking that during development of GECKO3 it is probably nice to have them in the GECKO repo, since we then don't need to change stuff in multiple repos and worry about having them synchronized. I do think it makes sense to then move them to another repo - my suggestion would be to place them in the GEM repos - so, the HumanGEMAdapter would go into the Human-GEM repo, etc. If someone makes a change, such as change how spontaneous reactions are stored in that repo, the adapter needs to be changed as well. One downside of that is that we cannot then (well, it would be weird) make YeastAdapter the default choice when calling functions, since GECKO will then not know about it. Technically it is still possible, but ...

[edkerk]

So, to explain how you access the functions in the adapter: function model = applyKcatConstraints(model,kcatSource,kcatComplex,kcatSubs,updateRxns,modelAdapter) %... params = modelAdapter.getSpontaneousRxns(model); %... end

Ok, what confused me is when you run adapter = Yeast8Adapter();, in MATLAB's Variables window adapter only contains params, but there is no (visual) indication that it also contains functions. But indeed, typing adapter. and clicking Tab also shows the function names (in Windows anyway).

[mihai-sysbio]

@johan-gson I think the previous commit a27b88c was also meant to address that the separator should be : instead of ;, so I provided specific suggestions for this. Or maybe I just got confused?

[johan-gson]

Ok, I made a final change to GECKOInstaller - I asked matlab for the separator instead of hardcoding ';' - I actually think the previous code only worked on Windows...

[mihai-sysbio]

Ok, I made a final change to GECKOInstaller - I asked matlab for the separator instead of hardcoding ';' - I actually think the previous code only worked on Windows...

Great improvements @johan-gson! I am happy to approve the PR, but I think it would be useful to get a second pair of eyes to look at it before merging.