Automated community model construction

[stelmo]

I have created a community model by stitching lots of iML1515 models together and while doing this found the reconstruction process to be slow and error prone, see #202 . Here are some wish list items that would've made it faster for me:

1. Automatic merging of models. This requires looking for external metabolites and turning the exchange-to-nothing reactions (normal exchange reactions) into exchange-with-the-environment reactions and adding an environment-exchange-to-nothing reaction for that metabolite.
2. Visualization of the connections (via environmental metabolites) between models - or at least some way of seeing if the exchanged fluxes/connections makes sense.
3. Functions to view the environmental fluxes
4. Functions to view the exchange fluxes of a specific microbe/model

I don't think this is too hard, but would be super nice to have...

[stelmo]

So maybe have a community model that looks like this:

struct CommunityModel
  S :: SparseMat
  b :: SparseVec
  c :: SparseVec
  lb :: SparseVec
  ub :: SparseVec
  mets :: Dict{String, Dict{String, Int}}
  rxns :: Dict{String, Dict{String, Int}}
end

The crucial difference between CoreModel and CommunityModel would the storage of the reaction and metabolite ids and how they map to reaction and metabolite indices in the model. I think something like this would make interpreting results easy for the user:

mets["ENV"] = Dict("env met 1" => 1, "env met 2" => 3, ... )
mets["organism a"] = Dict("met 1" => 2044, "met 2" => 4556, ... )
mets["organism b"] = Dict("met 1" => 8993, "met 2" => 10223, ... )

etc., and the same for the reactions.

Then, after running

vec = flux_balance_analysis_vec(community_model, Tulip.Optimizer)

The results could very easily be queried in dictionary format, e.g.

index = get(community_model.rxns["ENV"], "reaction i", nothing)
vec[index] # flux you care about

I suppose the user could also construct something like this themselves but it would be nice to generate these types of linked dictionaries during automated community model construction...

Another thought is that if there are 1_000_000_000s of reactions and metabolites lists of strings like in CoreModel, or worse StandardModel, they could get massive. However, most of the fluxes will never be queried after an analysis function, e.g. FBA, is run. Then it might be worth it to be able to specify which reactions get remembered by the model. E.g. environmental and species exchanges are important, but the internal reactions not so much...?

If only the environmental exchanges and interactions between (not within) microbes are stored then it would also be easy to construct a species map/visualization of the community, and maybe superimpose those fluxes on that.

[stelmo]

will be addressed by #270