EP Computation with CellNetAnalyzer

[Luca Schnepp-Pesch]

Hi everyone,

I’m currently working on computing extreme pathways (EPs) for genome-scale stoichiometric models and would like to use CellNetAnalyzer for this purpose. However, I’m unsure about the best way to proceed and what limitations exist in terms of network size and complexity, especially when the models include a biomass reaction.

I’d really appreciate any advice that could help clarify the feasibility and typical challenges involved and documentation on how to best use CellNetAnalyzer for EP computation.

Thanks in advance for your insights!

Best regards,
Luca Kristina

[Steffen]

Dear Luca Kristina,

I'm afraid that also CNA will not be able to compute the EPs in a genome-scale
network - but you might give it a try.

Proceed as follows (see also the more detailed descriptions in the manual):

0) Download and install CNA.

1) Import the model via the API function SBML CNAsbmlModel2MFNetwork
(for usage and options of API functions write "help <APIfunction>" in the command window).

2) Save the generated CNA project via the API function CNAsaveNetwork.

3) Use the API function CNAcomputeEFM to compute the EPs (importantly, you need to
set "convbasis_flag" argument to 1).

Hope this helps.

Best,
Steffen.

[Luca Schnepp-Pesch]

Dear Steffen,

Thank you very much for your quick and helpful response!

I have two follow-up questions:

1. SBML Preparation for EP Computation: Are there specific modifications required to make the SBML model suitable for computing EPs in CellNetAnalyzer? In particular, should all reversible reactions be split in forward and reverse, or only the internal reversible reactions?

2. Feasibility Estimate: Do you have a rough estimate of the network size limit for feasible EP computation in CNA? For example, would a model like the E. coli core with approximately 86 reactions, including around 40 reversible reactions and a dense biomass reaction, still be within a manageable range?

Thanks again for your support.

Best,
Luca Kristina

[Steffen]

Dear Luca Kristina,

Regarding your first question: you should first (re)consider whether you want to
compute (a) EFMs, (b) a convex basis (better: the generators of the solution space),
or (c) the EPs. As you probably know, the set of EPs is the convex basis of a
specifically reconfigured  network. As was demonstrated in this paper:
https://www.cell.com/trends/biotechnology/fulltext/S0167-7799(02)00034-3?large_figure=true

in almost all relevant applications, the EPs coincide with the EFMs and a reconfiguration is
thus not necessary (in fact, computing the EPs in the reconfigured network may then take
even longer than computing the EFMs directly in the original network, because of the 2-cycles,
which arise when splitting the reversible internal reactions and may complicate the algorithm).
Only if there are reversible exchange reactions, the EPs and EFMs would differ - but this case
is rarely relevant for realistic applications, as it would mean that an external compound is
considered to act simultaneously as substrate and product. Also, the interpretation of the
EPs in those cases can be problematic (see the reference above).

So are you really sure to compute the EPs and do they really differ from the EFMs, i.e.
does your network have reversible exchange reactions? [As another note: as was also
mentioned in the reference above, if the export/import of compunds follows the usual scheme
<--> Be <---> Bp<--->Bc  (the indices stand for extracellular, periplasmic and cytoplasmic),
then the EFMs and EPs will again coincide!].

If so, then you should first reconfigure the network (outside of CNA), import the model
via SBML and them compute the EPs as comvex basis in this network as described in my
previous post.

If you want to compute just the convex basis in the original network, do not reconfigure
the network (and set also here the convexbasis_flag=1 in the CNAcomputeEFM function).

Regarding your second question, it is really hard to estimate the feasibility or number of
EPs/EFMs since this will depend on many factors (in particular, on the dimension of the
nullspace, on the number of inputs/outputs, biomass reaction etc.).

Hope this helps,
Steffen.

[Luca Schnepp-Pesch]

Dear Steffen,

Thanks a lot for the detailed explanation, it really helped clarify things. Based on your advice, I’ll likely go with making all exchange fluxes irreversible and take advantage of the coincidence of meaningful EFMs and EPs.

Thanks again and best regards,
Luca Kristina