SBML test models

[matthiaskoenig]

Hi all, I just read the LASSIE paper and wanted to compare the results to libroadrunner I was involved with See: https://www.ncbi.nlm.nih.gov/pubmed/26085503 which achieves in my own experience ~5-10x speedup for large system compared to COPASI (which uses LSODA) and has no problem running reaction kinetics with arbitrary math on large systems (~3000 reactions and species).

I tried to run the provided 4096x4096 SBML model, but there seems to be an issue with the model equations because around ~10 seconds of simulation the concentrations run against infinity (i.e. step sizes cannot be fullfilled any more). Also before some of the concentrations have negative values, like X454, which than stops the integrator around 9.064 s See for instance X454 which runs against negative infinity shortly before the simulation aborts.

    time    [X454]
0   0   0.07
1   0.453   0.0426487581
2   0.906   0.0143706688
3   1.359   -0.0134872222
4   1.812   -0.0406795254
5   2.265   -0.0677183225
6   2.718   -0.0956023649
7   3.171   -0.1256590949
8   3.624   -0.1595044889
9   4.077   -0.1991152275
10  4.53    -0.2470183384
11  4.983   -0.3066406604
12  5.436   -0.3829256622
13  5.889   -0.4834716336
14  6.342   -0.6207917802
15  6.795   -0.8172666296
16  7.248   -1.1174850204
17  7.701   -1.625187808
18  8.154   -2.6505199791
19  8.607   -5.7315926781
20  9.06    -633.5137869673

In my opinion there should be no negative concentrations which seems to be an indicator for either wrong kinetics or initial values.

Are you sure that the kinetics are correctly encoded in the SBML file? I could imagine that the reason COPASI could not handle the file is that it is a non-integratable system (running to infinity) due to some encoding error of the ODEs. Could you check that the SBML has the correct kinetics? Could you provide a reference solution for the 4096x4096 SBML model simulated with LASSIE (do you see negative values)? Could you provide SBML files for the smaller examples in Table 1, i.e the 64x64 up to the 8192x8192 models, so I could run a full benchmark comparison of LASSIE against libroadrunner.

Thanks for any help Matthias

[matthiaskoenig]

Hi all, it looks like the SBML file has indeed some problems.

See discussion here https://github.com/sys-bio/roadrunner/issues/371

In L1V2 you cannot encode amounts of species, but all species are concentrations (which is not working with the kineticLaws you provide). You should encode the SBML file in a higher version of SBML, preferable L3V1 (or L2V4) which allows to define the species as amounts. Than you should also be able to simulate the model with cpu simulators for SBML like COPASI/roadrunner.

If you have any questions how to encode the model in SBML please let me know. I would recommend to write some python script (with libsbml python bindings) which converts your internal LASSIE format to SBML. You could than provide all the test files in SBML, so the performance could be compared to SBML based simulators.

The best Matthias

[aresio]

Dear Dr. König,

thank you very much for your kind observations and your interest in LASSIE. We are checking the SBML files and investigating the issue you pointed out. We will post our findings as soon as possible.

All the best, Marco

[matthiaskoenig]

Hi, any update on the issue? M

[aresio]

Dear Dr. König,

we apologize for the delay. We suspect that the problem was the software used to produce the SBML. This legacy code creates a model encoded using a very early version of SBML that might be partially incompatible with COPASI. We are currently developing a novel converter to investigate this possibility.

Marco

[matthiaskoenig]

Hi Marco, any update on this? M

On Wed, Jun 21, 2017 at 5:00 PM, Marco S. Nobile notifications@github.com wrote:

Dear Dr. König,

we apologize for the delay. We suspect that the problem was the software used to produce the SBML. This legacy code creates a model encoded using a very early version of SBML that might be partially incompatible with COPASI. We are currently developing a novel converter to investigate this possibility.

Marco

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/aresio/LASSIE/issues/1#issuecomment-310106749, or mute the thread https://github.com/notifications/unsubscribe-auth/AA29ut8Vu-8bBZMTv32kP-YPtuOZEuIYks5sGTAagaJpZM4NYDsa .

-- Dr. Matthias König Junior Group Leader LiSyM - Systems Medicine of the Liver Humboldt-University Berlin, Institute of Biology, Institute for Theoretical Biology https://www.livermetabolism.com konigmatt@googlemail.com Tel: +49 30 20938450 Tel: +49 176 81168480