Fix/removed beta-D-glucose and related reactions

[pranasag]

Main improvements in this PR:

Human-GEM had two definitions of cytosolic glucose and glucose-6-phosphate: glucose[c] and beta-glucose[c] (and respectively for the G6P). The conversion rxn beta-D-glucose[c] <=> glucose[c] (MAR07745) and the reactions with the "beta-" glucose, ending with fructose-6-phosphate[c] create unnecessarily alternative to glycolytic pathway from glucose[c] to fructose-6-phosphate[c], which has no added value to the model. Moreover, the "D-" and "unspecified" stereoisomer distinction is not there for any other sugar. Thus I suggest removing the respective metabolite species (MAM01388c and MAM01389c) and reactions (MAR07745,MAR07746, MAR07747, MAR07748, MAR07749).

I hereby confirm that I have:

• Tested my code on my own computer for running the model
• Selected develop as a target branch

[haowang-bioinfo]

@pranasag interesting propose.

The conversion rxn beta-D-glucose[c] <=> glucose[c] (MAR07745) and the reactions with the "beta-" glucose, ending with fructose-6-phosphate[c] create unnecessarily alternative to glycolytic pathway from glucose[c] to fructose-6-phosphate[c], which has no added value to the model.

May I know what evidence makes these reactions unnecessary to the model?

For example of reaction MAR07745, the catalytic gene ENSG00000143891 (GALM) is known to be expressed in kidney epithelium and many other tissues according to the Bgee database.

[pranasag]

Hi @Hao-Chalmers,
First, for all the reactions with beta-D-glucose[c] or beta-D-glucose-6-phosphate[c], there are concurrent reactions, using glucose[c] or glucose-6-phosphate[c], so the pathway remains there (and maintains a classical shape of glycolysis).
Second, I agree with your observation that mutarotases, such as GALM, are detected to be expressed in experiments and have a meaning there, because glucose in solution is in an equilibrium between the alpha- and beta rotational isomers (not sure whether this influences kinetic parameters of enzymes, though). However, neither mutarotation is relevant for reaction stoichiometry nor is defined completely in the model (then we'd have also a set of reactions for alpha-D-glucose as such, perhaps?). That'd be my rationale for removing this set of reactions.

By the way, I'm not sure why the YAML validation failed (says - reaction annotation not matching), I checked the changes introduced and these should be only related to removing 2 species and 5 reactions, as suggested.

[haowang-bioinfo]

@pranasag thanks for quick response.

By the way, I'm not sure why the YAML validation failed (says - reaction annotation not matching), I checked the changes introduced and these should be only related to removing 2 species and 5 reactions, as suggested.

The failure of YAML validation is due to the inconsistency between model (in YAML) and annotation files (reactions.tsv and metabolites.tsv). For example, if some reactions and/or metabolites are removed from model, the corresponding lines should also be moved from the annotation tsv files to deprecatedReactions.tsv and/or deprecatedMetabolites.tsv files under /data/deprecatedIdentifiers folder.

[pranasag]

@Hao-Chalmers my bad, I'll fix that in the next hours, thanks for the explanation!

[haowang-bioinfo]

@pranasag no worries.

However, neither mutarotation is relevant for reaction stoichiometry nor is defined completely in the model (then we'd have also a set of reactions for alpha-D-glucose as such, perhaps?).

This probably means that the model should be expanded with a complete set of "beta-" version reactions, instead of removing them?

[pranasag]

I would argue for removal rather than expansion, as (indicated above) the mutarotational state of glucose does not influence the stoichiometry of the reaction. Moreover, to my knowledge, other popular GEMs (at least ones I'm working with) do not have such distinctions because of this reason.

Added for the "do not have distinctions":
Only few models from BIGG database have additional species (such as this, this or this), while vast majority has only D-glucose.

(Please note that with the latest commit I also removed the gene ENSG00000143891 which becomes orphan when reaction MAR07745 is removed.)

[JonathanRob]

Thanks for your help, @pranasag.

I tend to agree with your logic, in that it doesn't really add anything meaningful to the model to have parallel reaction pathways for the different stereoisomers, since they don't produce anything different in the end (other than a few additional isomeric compounds). This is a bit similar to how we deal with lipid metabolism, where we don't try to model every possible combination of metabolites and reactions, but instead keep the most representative, common, or generic pathway form. Finally, having parallel reaction pathways can be problematic if simulations can seemingly compensate with the alternative pathway, when in reality, it may exhibit very poor efficiency.

On the other hand, I share @Hao-Chalmers's concerns of removing content/information from the model unless really necessary.

In this case, I would recommend removing the reactions as you suggest, @pranasag, except for the reaction transforming between the two glucose isomers (glucose and beta-glucose). This way, we still retain the information that this transformation can take place (and thus don't lose a gene association), but don't retain unnecessary components which are effectively a duplication of existing metabolites/reactions. I realize this creates a dead-end reaction and metabolite, but our approach with Human-GEM has been to allow dead-ends since they often contain valuable information and can easily be removed if not needed (e.g., for simulation).

What do you think?

[pranasag]

Good one, @JonathanRob, thanks for jumping in!
If creating dead-ends in favor of retaining information is still acceptable, I see your suggestion as a win/win option out of this. So I'll revert the commits and only remove beta-D-G6P metabolite and related reactions, leaving the conversion of glucose in the model.

[mihai-sysbio]

Only few models from BIGG database have additional species (such as this, this or this), while vast majority has only D-glucose.

@pranasag on a side note, I'm wondering is this might be an artifact of several models being based on the same template. Similarly, I see that the stereoisomers exist in Recon3D with a reference to HMR2.

[haowang-bioinfo]

I realize this creates a dead-end reaction and metabolite, but our approach with Human-GEM has been to allow dead-ends since they often contain valuable information and can easily be removed if not needed (e.g., for simulation).

@JonathanRob thanks for the clarification (this should be added to the Wiki as guidelines for future). Besides being a GEM, Human-GEM serves as a knowledgebase that allows dead-end reactions, which may be useful in other applications (e.g. integrative analysis).

According to a previous issue #44, unnecessary reactions should be initially deleted softly by blocking them ub=lb=0. While hard deletion will be implemented later in a major release (e.g. Human 2.0), this came from lessons learned from previous cases of first deleting and then adding back later. Should this soft deletion approach be applied here, what do you think? @pranasag

[JonathanRob]

Just to comment, @Hao-Chalmers, the "soft deletion" approach was developed before we had implemented the yaml workflow, when it was much harder to track and revert changes to the model (in particular deletion of components). Now that we have the convenient yaml-based workflow, I wouldn't see a need to use the soft deletion approach.

[haowang-bioinfo]

@pranasag I wonder if you also notice more beta- version metabolites (or stereoisomers) other than beta-D-glucose?

[pranasag]

@Hao-Chalmers there's no extra metabolites for fructose and mannose, but for galactose, there's galactose-1-phosphate[c] (MAM01911c) and alpha-D-galactose-1-phosphate[c] (MAM01322c). Some of the reactions are duplicated for these two species, but MAR08767 is unique to the alpha form, so, resolving this completely would require to remove reactions with alpha-D-galactose-1-phosphate[c] and to modify MAR08767 (substitute with MAM01911c).

[haowang-bioinfo]

Some of the reactions are duplicated for these two species, but MAR08767 is unique to the alpha form, so, resolving this completely would require to remove reactions with alpha-D-galactose-1-phosphate[c] and to modify MAR08767 (substitute with MAM01911c).

The reaction MAR08767 is associated with gene GALT (ENSG00000213930), which encode Galactose-1-phosphate uridylyltransferase (EC 2.7.7.12) that catalyzes the conversion of galactose-1-phosphate to glucose-1-phosphate in the Leloir pathway. According to UniProt, Rhea, MetaCyc and BRENDA, all the reactants and products are and only in alpha form.

In addition, there is no literature found by searching Pubmed with key words beta-D-galactose-1-phosphate and GALT. While two papers (271990 and 2011574) can be found by searching alpha-D-galactose-1-phosphate and GALT, both described the use of alpha form.

In sum, there is no evidence supporting the change of alpha form in MAR08767.

[pranasag]

That's really a broad overview of literature, thanks @Hao-Chalmers. Correct me if I'm wrong -- your conclusion is to preserve the two forms of galactose-1-phosphate in the model. If that's so, then I'd appeal to what @JonathanRob said about having same end-products after the pathway branches to another metabolite species (being the same chemical thing), in this case, between the galactose-1-phosphate[c] and alpha-D-galactose-1-phosphate[c].
It seemed like that removing the additional glycolysis detour (the initial PR) was sort-of resolved, what's now the consensus on this one? Pinging, again, @JonathanRob and others.

[JonathanRob]

I'm not sure I fully understand what the consensus is at this point. Would this interpretation be correct, or did I misunderstand?

1. Remove glycolysis detour (except for MAR07745 which converts between alpha and beta glucose)
2. Leave the galactose pathway as-is

[pranasag]

@JonathanRob even though I'd argue that the issue I see with the galactose pathway is very similar to the one for glycolysis (from the modelling perspective, while agreeing to the experimentally-relevant observations @Hao-Chalmers collected), maybe then let's stick to clearing up glycolysis at this time?

[haowang-bioinfo]

I'm not sure I fully understand what the consensus is at this point. Would this interpretation be correct, or did I misunderstand?

1. Remove glycolysis detour (except for MAR07745 which converts between alpha and beta glucose)
2. Leave the galactose pathway as-is

@JonathanRob makes sense to me

let's stick to clearing up glycolysis at this time?

@pranasag agree

[haowang-bioinfo]

Revised implementation plan after GPR checking:

• Remove beta- form metabolites MAM01388c and MAM01389c
• MAR07745: retain as-is
• MAR07746: remove, and modify GPR of its counterpart (MAR04394) by taking away ADPGK (which is ADP-dependent)
• MAR07747: retain, but change from beta- form to generic
• MAR07748: remove, seems no evidence supporting this conversion reaction
• MAR07749: remove, duplicate to MAR04381

@pranasag @JonathanRob what do you think?

[pranasag]

Well done @Hao-Chalmers, sounds great! Also thanks for noticing MAR07747 being ADP-dependent, I accidentally overlooked this.

[JonathanRob]

Looks great, @Hao-Chalmers

[haowang-bioinfo]

@pranasag how about restarting a new PR to implement the revised plan, this probably is easier than making amendments to the current one?

Instead of pushing from a forked repo, I suggest making new PR directly to Human-GEM (you've assigned with write access), so that we can help in case there are issues raised by GitHub actions.

[pranasag]

Thanks @Hao-Chalmers, I'll make a branch, do the changes and submit a new PR then.

[mihai-sysbio]

@Hao-Chalmers what was the reason to keep MAM01388c in the model in the revised plan?

[JonathanRob]

@mihai-sysbio MAM01388c (beta-D-glucose) is part of the conversion reaction MAR07745, which we wanted to keep because it represents unique chemistry that is not duplicated elsewhere. Removing this would remove information (and a gene association) from the model.