Re: [sbml-qual] Design choices for SBML Logic Models

[Haoran Yu <hal...@gm...>]

Sorry for not attaching the links to the Figures, here are the links in
case the Figures don't shown up in the previous email.
Figure 1
<https://1.bp.blogspot.com/-fs90lVuf2xw/WVypEN9zJGI/AAAAAAAAGfY/gsEbiGH228w9ocY1hgg0v7cDlwxpkllQgCLcBGAs/s320/or-simple.png>
Figure 2
<https://2.bp.blogspot.com/-u_5AitRQqfw/WVypIP5Rz0I/AAAAAAAAGfc/CeQGeo3Ua6gOmDqT8KJeuvbj_-ze3_ddQCLcBGAs/s1600/or-simple-new.png>
Figure 3
<https://4.bp.blogspot.com/-gmKV5sxzvbs/WWhtqdAzU_I/AAAAAAAAGiY/Rx1m2l5Z9Gw5aXIUmJn9tdg75udVW-R_wCLcBGAs/s1600/compartment.png>

Thank you again!
Haoran

On Fri, Jul 14, 2017 at 2:55 AM, Haoran Yu <hal...@gm...> wrote:

> Hi all,
>
> I have a couple of questions regarding how to design Logic Models when
> writing SBML from scratch. The questions might not be specifically for SBML
> 'qual', but I thought it might be convenient to have all my questions in
> one place rather than across different mailing lists.
>
> I'm currently implementing the following bidirectional converters between
> SBML and SBGN-ML:
>
> (1) SBML 'core'+'layout'+'render' <-> SBGN-ML PD
> (2) SBML 'core'+'qual' <-> SBGN-ML AF
>
> I have made some progress in both converters (1) and (2). All the
> questions I have is for the SBGN - > SBML conversion. The questions were
> posted on my blog over the last 2 weeks, Week 10
> <http://haleyyew-gsoc.blogspot.ca/2017/07/week-10.html> (section 2) and Week
> 11 <http://haleyyew-gsoc.blogspot.ca/2017/07/week-11.html> (section 1.1).
>
> The SBGN examples that motivated the questions are
> libsbgn/test-files/PD/or.sbgn
> <https://github.com/sbgn/libsbgn/blob/master/test-files/PD/or-simple.png>
> and libsbgn/test-files/AF/compartment.sbgn
> <https://github.com/sbgn/libsbgn/blob/master/test-files/AF/compartment.png>
>
> *Question 1:*
> There are 2 alternative mappings for SBGN PD->SBML for the example
> "or.sbgn" that involves choosing between 'layout' SpeciesGlyphs,
> ReactionGlyphs and GeneralGlyphs.
> I included a color coded diagram to illustrate the 2 mapping alternatives, Blue
> is SpeciesGlyphs, Green is ReactionGlyphs and Red is GeneralGlyphs.
> [Figure 1]
> [Figure 2]
>
> In Figure 1, I mapped the 'OR' SBGN Glyph to a ReactionGlyph ,and
> therefore all Arcs connecting the 'OR' glyph should become
> SpeciesReferences. But I found this mapping to be inadequate. Arc [1] is a
> ModifierSpeciesReference, and by *Rule 21111* in the
> *sbml-level-3-version-1-core-release-2* specification, the *species *attribute
> must refer to an existing Species. Which means Arc [1] refers to the 'OR'
> glyph and 'OR' must be a Species, otherwise Arc [1] can't be a
> ModifierSpeciesReference of the green reaction. That is why I had to make
> Arc [1] a GeneralGlyph instead without associating with any core Model
> components.
>
> In Figure 2, the reason that the 'OR' SBGN Glyph is changed from a
> ReactionGlyph to a SpeciesGlyph is to fix the problem with the Arc [1] in
> Figure 1. The 2 Logic Arcs below *Enzyme1 *and *Enzyme2 *become
> GeneralGlyphs, and the 'OR' SBGN Glyph becomes a Species. Then Arc [1] can
> be a ModifierSpeciesReference in the green reaction below, and associate
> with the 'OR' Species. The 2 Logic Arcs should be GeneralGlyphs because
> they are not real Model components anyway.
>
> I have implemented the mapping in Figure 1, but later changed the
> implementation to the mapping in Figure 2. *Since neither of the 2
> alternatives are perfect, I'm wondering if there is a third alternative
> that might be a better mapping? Or any suggestions that improve the
> mappings in either Figure 1 or Figure 2? *
>
>
> *Question 2:*
> While working on converter (2), the "compartment.sbgn" example in SBGN AF
> is more complex than I thought, and I have trouble mapping it to SBML
> 'qual'. I can map an SBGN Logic Arc to a SBML 'qual' Transition, where the
> Input and Output QualitativeSpecies could be anything connected to the
> Logic Arc.
> [Figure 3]
> ​
> But in the more complex case where many Logic Arcs meet with a Logic
> Operator, and then output an Arc to connect to the target Glyph. All of
> these Arcs and Glyphs would belong to a single Transition in 'qual'. Each
> Logic Arc has an Input. Then each of them would have a separate
> FunctionTerm. But each FunctionTerm must have a resultLevel. The problem is
> that in Logic Models, the possible Input and Ouput values could only be 0
> or 1. *Are there are only 2 possible FunctionTerms in all Transitions
> (resultLevel 0 or 1)?*
>
> Due to my current choice mapping: the 'AND' glyph would be mapped to a
> QualitativeSpecies, and for each SBGN Glyph that is connected to a Logic
> Arc, I map the Glyph to a QualitativeSpecies. Then I create SpeciesGlyphs
> that correspond to the QualitativeSpecies. But the 'AND' Logic Operator is
> part of a Transition, and the Arc that acts as the Output must connect to
> only one QualitativeSpecies. Therefore the 'AND' glyph does not have a
> mapping, i.e. it is lost in the mapped SBML Model.
>
> *Is there a better mapping that is able to preserve the 'AND' glyph? *
> I have thought about storing the 'AND' as part of the Math element in a
> FunctionTerm. But I have trouble coming up with a good algorithm to map the
> SBGN Glyph+Arcs to a Transition containing multiple FunctionTerms. It would
> be great if someone could help me trace an algorithm for mapping of the
> "compartment.sbgn" example.
>
>
> Thank you for all your help!
> Haoran
>
>