cooperative.tradeoff VS grow()

[rebminso]

Hello @cdiener, I have been trying to use MICOM in my research and I have a few queries regarding different approaches to defining growth rates.
Specifically, I am trying to understand whether there is a difference in the taxon growth rates obtained from the following two approaches.

• data is a data table with multiple community information
  manifest = build(taxonomy = data , model_db = None , out_folder = 'allsample', threads = 8, solver = 'cplex')

1. grow(tradeoff=1.0)

res = grow(manifest, model_folder="allsample", medium=medium, tradeoff=1, threads=8)
res.growth_rates
*allsample is a folder containing pickle files

• 1.1 tradeoff=0.8 (all taxa grow at this tradeoff)

2. cooperative_tradeoff(fraction=1.0)

com = load_pickle("file.pickle") # one of the pickle file
sol = com.cooperative_tradeoff(fluxes=True, pfba=True)
sol.members

• 2.1 fraction=0.8

In a nutshell:

1. Are the grow() and cooperative_tradeoff() approaches expected to give the same taxon-level growth rates for the same community? If so, which method should be used to obtain more biologically realistic taxon growth rates?
2. Community growth rate is usually obtained using cooperative_tradeoff(). Is it also possible to obtain the community growth rate using grow() or I'm missing out on something?
3. When comparing taxon-level growth rates across samples, is it appropriate to directly compare the values, or is some form of normalization recommended?

Any insights or clarification would be much appreciated. Thank you very much for your time and for developing this tool!

[cdiener]

Hi @rebminso,

grow uses cooperative tradeoff FBA as well under the hood. fraction and tradeoff describe the same parameter. It's just unfortunate naming I should fix. They will give the same results for growth rates within solver accuracy.

1. Those are just different APIs. grow is part of the workflows and supposed to be used for multiple samples. It sets sensible defaults but has less flexibility. cooperative_tradeoff is part of the single model API. It does not have as many checks and requires manual control for setting the medium, finding minimum imports etc. But it does give more control for custom strategies.

2. Yes it's possible. It's just sum(abundance * growth rates) for each sample in the growth rates output. It's planned to be included as a helper in GrowthResults. If every sample is simulated in the same medium those tend to be very close though.

3. If you used the same medium for every sample then yes. They can be compared directly. If not I would normalize by the nutrient content of the medium maybe to account for the bias that richer media will lead to higher growth rates.