Reactions are unaware of groups (and subsystems)

[akaviaLab]

trafficstars

Problem description

The model has groups that contain members (list of reactions). However, the reactions do not have a groups field, and the subsystems is always an empty string. I think it should behave like genes, where reaction has a set of groups and relevant functions _associate_group(self, cobra_group) _dissociate_group(self, cobra_group)

If that's approved, models that include groups should associate them with reactions when loaded.

This is similar, but different from #543 and #473 - I understand this should be encoded as groups, but I think reactions should know their groups.

I would suggest having subsystems either be

1. Deprecated
2. Just a link to groups, basically reaction.subsystem = reaction.groups
3. A list of textual group IDs and not full group members (similar idea to GPR string vs genes)

If the key developers can comment on this.

[gregmedlock]

I think the main implementation concern is being careful with circular references as much as possible so that deletion and addition of objects doesn't mangle group membership. This is probably why _associate_gene and _dissociate_gene are written as "private" methods.

I think this would be a welcome feature but would require a lot of work on the API (and might seem weird if functions for associating genes with reactions are still private). Implementing it in Object might be a good option so that any of the cobrapy objects can be associated/dissociated with a group via this method.

[akaviaLab]

I'm a bit confused. I wasn't planning to use _associate_gene, just make an equivalent for _associate_group, which I guess will be private as well. Are entities other than reaction associated with groups?

[cdiener]

It's true that Reaction.subsystem is a relic of SBML 2 times. We had a lot of bugs because of multiple references as @gregmedlock mentioned when we introduced the context managers. This is why I pushed for a "single source of truth" model where only one entity tracks connections between objects. A metabolite does not know what reactions it belongs to. Metabolite.reaction actually goes through all reactions and returns the ones that the metabolite belongs to. That is a bit slower but way more stable in terms of maintenance.

It would be pretty easy though to do the same thing for Reaction.groups and I think that would be a good contribution.

[matthiaskoenig]

Quick input:

• yes, other objects then reactions are in groups, in the context of cobrapy the relevant objects are reaction, species, gene, compartment (and could also be a GPR).
• single objects can be part of many groups, i.e. a species like atp could be in many groups.
• groups can be hierarchical, i.e. groups could contain other groups

Groups are much more powerful then the subsystems, and the information in subsystems is only a subset of the possible group information.