feat: I/O of SBO terms from SBML file

[edkerk]

Description of the issue:

• RAVEN is not able to I/O SBO terms that are specified in SBML files.

Reproducing this issue:

SBO terms are included in the SBML file as following (similar for metabolites and reactions):

<species boundaryCondition="false" constant="false" ...
   hasOnlySubstanceUnits="false" id="M_s_0001" name="(1-3)-beta-D-glucan" ...
   metaid="M_s_0001" sboTerm="SBO:0000247" compartment="ce" ...
   fbc:charge="0" fbc:chemicalFormula="C6H10O5">

When loaded in Matlab using TranslateSBML, SBO terms are easily accessible at (similar for metabolites and reactions):

modelSBML.species(:).sboTerm

Parsing this in exportModel and importModel should be relatively straight forward, and should be stored in the metMiriams and rxnMiriams structures.

Note that ravenCobraWrapper does already have the option to parse SBO terms from the MIRIAMS structures to their COBRA counterpart fields (metSBOTerms).

System information

1. RAVEN version: all
2. Operating system: all

I hereby confirm that I have:

• Followed the guidelines to install RAVEN.
• Checked that a similar issue does not already exist

[haowang-bioinfo]

should be stored in the metMiriams and rxnMiriams structures.

COBRA deals with SBO terms in a different way by using metSBOTerms and rxnSBOTerms fields (according to yeast-GEM v8.3.3). Since there is an ongoing issue toward aligning RAVEN and COBRA model fields (#184), maybe good to have a discussion there about weather storing in xyzMiriams or separate fields before the implementation.

[edkerk]

@Hao-Chalmers For merging the model fields, it was proposed to use the COBRA version, taking advantage of their new getModelFieldsForType and similar functionality. So solving this issue would indeed not be relevant for future versions of RAVEN that directly work on COBRA models, but it would still be good to fix this for the meanwhile for the classical RAVEN model format. It's not so hard to code, will dedicate an hour to before the end of the year.

[JonathanRob]

This feature is currently of interest to me, so I implemented the changes into exportModel and importModel (now on feat/SBO_SBML_IO branch).

Regarding changes to importModel, I originally planned to parse SBO terms from the .sboTerm fields of the SBML file (for metabolites, reactions, genes, and compartments), and add them to the corresponding miriam fields in the model structure. However, this leads to a situation where if a user exports a model (with no SBO terms specified in any miriam fields), then re-imports that SBML, the model will now have the default SBO terms (assigned in exportModel) loaded into all the miriam fields.

This seemed unnecessary to me, so I left the original formulation which will only import SBO terms from the SBML if it is part of the .annotation field (i.e., only if the SBO term was included in the miriam field when exported). This results in miriam fields that remain identical after export/reimport cycle, which I think is preferred.

I also have a quick question: The default SBO term assigned to genes by exportModel is SBO:0000252, which corresponds to "polypeptide chain". Is there a reason we are not using SBO:0000243, which corresponds to "gene"?

[simas232]

@JonathanRob, feel three to change the default SBO terms in exportModel to the most relevant ones. When I implemented OutputSBML into exportModel two years ago, I think I used SBO terms which were considered by COBRA developers during that time.

[edkerk]

Perhaps check if particular SBO terms are part of a test in memote. If we'd like to use different SBO terms, we should then also advocate this with memote people (and COBRA?).

Regarding the I/O function you introduced, I'll check it out. I think it's also important that it can deal with non-RAVEN models, so where do COBRA- and cobrapy-exported models store the SBO terms in the XML file?

[JonathanRob]

@edkerk Ah, I should have mentioned, memote uses SBO:0000243 for genes, which is what I am suggesting to change it to.

But good point regarding functionality with COBRA-exported models, I will probably need to add a few things for that.

[edkerk]

@JonathanRob In that case, definitely use SBO:0000243 for genes!

[simas232]

@JonathanRob, regarding SBO terms parsing in importModel, we may add the check for uniqueness of SBO terms for each entity type separately. For instance, if there are at least two unique SBO terms between all metabolites, we may consider importing all SBO terms into metMiriams field. However, if the same SBO terms are used for all entities of the same type (e.g. genes), we may neglect them, since they would be added again to the model upon exportModel.

On the other hand, if we agree to implement this check, the changes would be made to exportModel as well. That is, there would be a check, whether xyzMiriams contains SBO terms, and if it does, then it would neglect the default SBO terms and use the ones available in xyzMiriams. But in such case, the written SBML file would have duplicate information (SBO terms in .annotation and .sboTerm), perhaps not the best idea after all.. Unless we do not write SBO terms to .annotation during the model export.

[edkerk]

ravenCobraWrapper should also be updated to parse between geneSBOterms and geneMiriams. For rxns and mets it should already be working.

[JonathanRob]

@simas232 I think a check for uniqueness of SBO terms when importing could be a good option, since the user probably doesn't really need the terms if they're identical for every entry of a particular entity.

As for the exportModel aspect, this check is essentially already implemented; i.e., the list of all SBO terms is first populated with the default values, and they are overwritten if an SBO term is found in a corresponding xyzMiriams field.

Regarding the duplication of SBO term information in the SBML file (present in both .annotation and .sboTerm fields), maybe it would be better to remove it from the .annotation fields altogether. It seems that Cobra-style models anyway use .sboTerm instead of .annotation.

[simas232]

Nice! Okay, then we should try to implement these changes to importModel and exportModel accordingly and finally solve this issue.

[JonathanRob]

• exportModel now only adds SBO terms to the SBML .sboTerm fields, instead of duplicating the information in the .annotation fields. I have also updated the default gene SBO term to SBO:0000243, which is what memote uses.
• importModel now reads SBO terms from the SBML .sboTerm fields, which should be compatible with both RAVEN- and COBRA-generated SBML files. Note that SBO terms are not imported if they are all identical for a given entity (e.g., if all metabolites have the same SBO term, it will not be imported into the model structure).
• ravenCobraWrapper now converts gene SBO terms between the .geneMiriams and .geneSBOTerms fields of RAVEN and COBRA model structures, respectively.

I'll soon make a PR to implement these changes.