lcsb-biocore / fbcmodeltests.jl Goto Github PK

Some models (e.g. community models, older models) employ split biomass functions that will fail most of the biomass function tests. Make the testing procedure robust against them.

Currently PTS style transporters are not recognized.

Right now any missing information is causes all metabolites to be marked as duplicated. This gives too many false positives.

Thanks for putting together these test suites, very nice work!
I have some usability questions (possibly issues):

A) Minor: it's unclear what model formats are accepted, since it's just a string of the model file. The example points to json, which is not what I would have expected.

B) Minor: it's unclear to me why there is the need to have 3 different docker images instead of one. As a user, I'd prefer to have one image that includes all everything (including dependencies), and pass different commands to it with docker run.

C) Major: I can't seem to run the Memote-style test on yeast-GEM:

FBCModelTests.Memote.generate_report("yeast-GEM.mat", Tulip.Optimizer)

since it ends up in an infinite loop that prints:

^C┌ Warning: Increase regularizations to 1.4901161193847656e-6
└ @ Tulip ~/.julia/packages/Tulip/SRdil/src/IPM/HSD/step.jl:47

This issue is used to trigger TagBot; feel free to unsubscribe.

If you haven't already, you should update your TagBot.yml to include issue comment triggers.
Please see this post on Discourse for instructions and more details.

If you'd like for me to do this for you, comment TagBot fix on this issue.
I'll open a PR within a few hours, please be patient!

This should get fixed once LCSB-BioCore/COBREXA.jl#785 is addressed.

With COBREXA 2.0 parse_annotations becomes unnecessary (see LCSB-BioCore/COBREXA.jl#788).

Right now many tests call fva internally. This is quite wasteful since it is an expensive operation. If possible, rewrite those tests to share the results of a single fva.

• metabolite connectivity
• unbounded flux in default medium

Currently looks_like is used to identify biomass and exchange reactions (usually not balanced). This is not robust, SBO terms should be used instead. Functions like is_biomass_reaction and is_exchange_reaction implement SBO matching in COBREXA, use that here.

While trying to run fbcmodeltests-run-frog via Docker on Human-GEM, it took way too many hours to compute with the default settings, so I tried for a little to manually add workers and failed. It would be great to allow passing of parallelization params to run-frog, since the underlying function FROG.generate_report already has support for it.

Currently the reaction annotation conformity does not work as expected (compare to output from memote.py). I think something is wrong with the regex expressions.

something like this:
https://github.com/LCSB-BioCore/SBML.jl/blob/9a674d010b36336cc843040c388693d0302f5c4d/test/loaddynamicmodels.jl#L66-L73
except throwing outright error on hash mismatch

...Perhaps there could also be a nice commandline utility for this purpose. :D we already have the docker in #22 . :D

Also good for CIs.

1. Test if genes, metabolites and reactions each have the correct annotations.
2. Improve the transporter tests (see LCSB-BioCore/COBREXA.jl#786)

Many of the config structs have dictionaries like

Dict(
        "trp__L" => "trp__L_c", 
        "cys__L" => "cys__L_c",
...
)

that map standard ids to namespace specific ids (in this case bigg). Change this to something like

Dict(
        "trp__L" => ["trp__L_c", "MNXM741553", "chebi:57912", ...], 
...
)

and modify the checking functions to make the default behavior more robust.

Let's implement all the memote tests:

• Consistency
• Basic information
• Metabolite information
• Reaction Information
• GPR associations
• Biomass
• Energy
• Network topology
• Matrix conditioning
• Reaction annotations
• Gene annotations
• Metabolite annotations

instead of

mutable struct someconfig
  x :: Int
end

default_someconfig = someconfig(1234)

do

Base.@kwdef mutable struct someconfig
  x :: Int = 1234
end

...this also makes seriously nice constructors, so that users can do:

someconfig(some_particular_nondefault_variable=4312)

...so that curators can check stuff using lower tolerance, such as

fbcmt-compare-frog report1 report2 --absolute-tolerance 1e-5 --relative-tolerance 1e-3

cc @ntung @rsmsheriff