@pstjohn check out https://github.com/biosustain/cameo/blob/devel/cameo/flux_analysis/structural.py,
MinimalCutSetsEnumerator and ShortestElementaryFluxModes. Sorry no documentation yet. But this unit test probably helps https://github.com/biosustain/cameo/blob/6babaffb50ea1c5f342fec255574f36e9350fa0e/tests/test_flux_analysis.py#L303. This is really experimental though. We were planning to use this for http://journals.plos.org/ploscompbiol/article?id=10.1371%2Fjournal.pcbi.1003378 but so far we have not got it completely right I think (need to talk Kristian). I think the application of real EFM software is fairly limited in both cobrapy and cameo. Calculating the shortest EFMs (like described in the paper) has some value though. You mentioned that you had nice wrappers for some of the better EFM tools in your cobrapy fork. Why not just make nice separate packages since they come with some heavy dependencies? I think we should try to avoid making cobrapy larger. With Python packaging being slightly less traumatic now, I think it makes sense to have many packages that do certain things very well. Of course we could highlight those on http://opencobra.github.io/.
@pstjohn Our implementation of ShortestElementaryFluxModes is based on the Von Kamp & Klamt paper that Niko linked to above. You can find the algorithms in the methods section. Currently we use ALGO2 for cplex and ALGO1 for other solvers. Finding elementary flux modes is very straight forward. You instantiate a flux mode enumerator by
ShortestElementaryFluxModes(model) and iterate over this object. This will return all flux modes ordered by length. I'm fairly certain this method in cameo works correctly. The MinimalCutSetsEnumerator is based on http://bioinformatics.oxfordjournals.org/content/28/3/381.short , but there still seems to be some issues with our implementation
@phantomas1234 , splitting the packages probably makes sense. I'll reach out to the original authors and make sure they're OK with it. @KristianJensen , thanks for the description! In case you were wondering, I've had good results using the
efmtool from marco terzer (doi:10.1093/bioinformatics/btn401) for EFMs and the berge algorithm from Christian Jungreuthmayer (http://www.biomedcentral.com/1471-2105/14/318) for minimum cut sets. Java / C++, respectively, though.
2 quick Qs: is the pathway predictor based on GEM-path? does the optknock code use the optknock procedure within cobrapy, or was it rewritten?
github is down (!! :worried: ) or I would just check out the code
We force a little bit of the compound to be produced m
Optknock was rewritten. We add an integer cut to yield multiple solutions
My advice is, do not use optknock cause it's slow and doesn't scale
You can run optgene with reactions or genes as targets. With FBA you can try 20000 designs in 5 min and you will get the 50 top solutions
@zakandrewking https://biosustain.github.io/cell-factory-design-course/ might be helpful with the teaching.
Can you help about my question?
OptGene is just a high-level wrapper for GeneKnockoutOptimization that stays close to the original publication, GeneKnockoutOptimization can easily be customized, one could for example swap the underlying heuristic from genetic algorithm to particle swarm or whatever etc.
Also cool that you started using it 👍please let us know if we can help you with something or you encounter any issues. We're close to submitting a preprint to biorxiv so feedback is very valuable.
I'm currently migrating a set of pipelines from Java and MATLAB (using OptFlux and COBRA Toolbox) to Python with cameo. Using the same solver, some of these are running too slowly with cameo (even across 128 CPU cores), but I still need to do some profiling... I'll let you know once I have investigated further.
I would be very much interested in the outcome of your profiling. Something to consider in terms of speed is to avoid copies of models at all costs. Instead you can use TimeMachine (please look in the cameo code for examples) to roll back changes to a model. You might also take a look at the latest cobrapy release where you can just write
with model: ...(like here https://twitter.com/DDDeCaF/status/841974326526128128). The latest cobrapy release also uses optlang. We're going to migrate cameo soon to the latest cobrapy release too.
First, how does one define the optimization objective in OptGene? In OptFlux one can choose between product yield or BPCY. In cameo I don't see that option.
Does cameo allow visualization of fluxes in customized escher maps ?
Can one add reactions to cameo models in a single line? Like in model.add("a + b -> c") ?
@joaocardoso Niko says you're the expert :)
Can one add reactions to cameo models in a single line? Like in model.add("a + b -> c") ?
You can build a reaction from a string and then add that to your model. https://cobrapy.readthedocs.io/en/latest/cobra.core.html#cobra.core.reaction.Reaction.build_reaction_from_string
Can one export models in JSON format ?
https://cobrapy.readthedocs.io/en/latest/cobra.io.html#cobra.io.json.to_json
by the way, you guys could add model.add_reaction_from_str like we have in framed, i find it super convenient
Objective functions in cameo can be found at
cameo.strain_design.heuristic.evolutionary.objective_functions