BioPAX, SBML, SBPAX and COMBINE (was: Pathway Databases with Quantitative Data?)
Oliver Ruebenacker
On Tue, May 24, 2011 at 12:19 PM, Emek Demir <de...@cbio.mskcc.org> wrote:
> Also I would like to see some discussion how we will avoid use-case overlap
> with SBML in the spirit of Combine. Some quantitative data like equilibrium
> constants are semi-universal. Concentration on the other hand is not. There
> is little benefit of exchanging concentrations unless you want to close your
> system and use it for simulations - like SBML. If you want to think about
> the open data aggregation and integration use case it makes little sense. In
> fact it might be inhibitory. I have strong reservations against taking the
> latter path. Most of the data that is collected in Molecule Pages, AFAIK,
> falls into the first category so it might be a good point to start
> considering what to include and what not.
The typical case goes something like this: a pathway database has
data that perfectly falls into what BioPAX is intended for, so they
export to BioPAX (notably, most export to SBML, too, but omit all the
rate laws and initial conditions). Now they have some quantitative
data and they want to export these, too. A few points of quantitative
data does not make it a model, so it should still be the domain of
BioPAX. But once there are enough such data points, it may imply a
model.
Where would the boundary between BioPAX and SBML be, if there is
one? BioPAX already has a few quantities, i.e. equilibrium constant
and free energy, with parameters like temperature, pH and pMg. And, of
course, stoichiometric coefficients. In SBML, there is an effort to
support constraint-based modeling, where some results can be obtained
with stoichiometric coefficients as the only numbers needed. They
could readily use BioPAX.
A concentration is usually scenario-specific. But some
concentrations are stable enough to be considered universal. Where you
would draw the line depends on the precision in which you are
thinking. A range of possible (or healthy or wild type) concentrations
would be a universal statement. What about a 'typical' concentration,
would that be universal or specific? The line between universal and
specific is necessarily blurred, since the universal is the set of all
possible particulars and can only be learned through the particulars.
That's why BioPAX talks about experimental evidence.
To demonstrate how similar BioPAX and SBML are, let me quote some
core definition, but replace some typical terms that would give away
whether its BioPAX or SBML by more neutral terms. Let STUFF be a
physical entity or species, ENCOUNTER an interaction or process,
CHANGE a conversion or reaction, PLACE a cellular location or
compartment and let SOLID mean either 'physical' or 'chemical' and
NETWORK either 'pathway' or 'model' and PEOPLE be biologists or
modelers. Can every one immediately say which of the following
definitions are BioPAX and which are SBML?
"A pool of entities, where each entity has a SOLID structure."
"A pool of entities that (a) are considered indistinguishable from
each other for the purposes of the NETWORK, (b) may participate in
CHANGES, and (c) are located in a specific PLACE."
"Any kind of ENCOUNTER that can change the quantity of one or more
STUFF in a NETWORK."
"A ENCOUNTER in which one or more SOLID entities is SOLIDLY
transformed into one or more other ones. [...]"
"A set or series of ENCOUNTERS, often forming a network, which
PEOPLE have found useful to group together [...]. NETWORKS can have
SUBNETWORKS."
Do you think BioPAX or SBML would significantly change of we
switched the definitions?
The idea often floated that the COMBINE formats are (or should be)
non-overlapping makes sense in some cases. For example, graphical
notation as such is probably orthogonal to everything else (in a
sense). But BioPAX and SBML are conceptually too close to each other.
I don't think it is possible to find consensus on a dividing line
between BioPAX and SBML, let alone enforce such a line, given the
decentralized development of extensions and packages.
It makes more sense to me to go the other way round: acknowledge
that there is not only an overlap, but a close relationship between
BioPAX and SBML and define it, i.e. specify a common core of BioPAX
and SBML (maybe COMBINE Core, or COCORE). I mean, currently, we are
saying that we have the species glucose in the cytosol which is
annotated by a MIRIAM qualifier that says that this is the same as the
physical entity glucose in the cytosol. Why don't we just say this is
glucose in the cytosol once? Or consider complexes: should we use the
qualifier "has part" to point to a component, or should we use the
qualifier "is" pointing to a complex which then has a component, or
both? Why not have one unified way.
By the way, why don't we build a COMBINE infrastructure? How about
having two mailing lists, one for all people and one for all COMBINE
editors, team members and advisers? I would be happy to open two
Google groups, if there is interest.
Take care
Oliver
--
Oliver Ruebenacker, Computational Cell Biologist
Systems Biology Linker at Virtual Cell (http://vcell.org/sybil)
Turning Knowledge Data into Models
Center for Cell Analysis and Modeling
http://www.oliver.curiousworld.org
Emek Demir
> Where would the boundary between BioPAX and SBML be, if there is
> one? BioPAX already has a few quantities, i.e. equilibrium constant
> and free energy, with parameters like temperature, pH and pMg. And, of
> course, stoichiometric coefficients.
law of mass conservation. Eq. constants and free energy are not unless
they are specified as formulas where effecting concentrations are
variables.
> A concentration is usually scenario-specific. But some
> concentrations are stable enough to be considered universal.
Example? Stable enough under which context? Can you represent those
contexts in BioPAX with existing mechanisms? Or do we need to introduce
a finer/more detailed context representation?
> Where you
> would draw the line depends on the precision in which you are
> thinking. A range of possible (or healthy or wild type) concentrations
> would be a universal statement. What about a 'typical' concentration,
> would that be universal or specific?
An average, specified together with a variance under some distribution
model would be a universal -yes..
> The line between universal and
> specific is necessarily blurred, since the universal is the set of all
> possible particulars
"All swans are white" is a universal statement and it something very
different than its existential counterpart "I observed a white swan". To
relate one to the other you need some philosophical structure - you
can't simply call it a set.This is something often stated in logic
theories, so from that perspective you are right. But in terms
philosophy of science, this is an empiricist view most recently argued
by logical positivists and is considered passé. Popper demolished it in
The Logic of Scientific Discovery. More accepted modern epistemological
views (such as those by Popper, Kuhn or their successors) clearly
separate a universal from an existential. Some of the conflicts that I
see between OWL's own semantics and science stems from this difference.
> and can only be learned through the particulars.
> That's why BioPAX talks about experimental evidence.
Umm.. yes. But BioPAX do not store particulars.. it just points to the
instances that was used to derive this universal statement. It is there
to be used purely to determine how much you are going to trust that
statement. But what you are suggesting is something categorically different.
> To demonstrate how similar BioPAX and SBML are, let me quote some
> core definition, but replace some typical terms that would give away
> whether its BioPAX or SBML by more neutral terms. Let STUFF be a
> physical entity or species, ENCOUNTER an interaction or process,
> CHANGE a conversion or reaction, PLACE a cellular location or
> compartment and let SOLID mean either 'physical' or 'chemical' and
> NETWORK either 'pathway' or 'model' and PEOPLE be biologists or
> modelers. Can every one immediately say which of the following
> definitions are BioPAX and which are SBML?
>
> "A pool of entities, where each entity has a SOLID structure."
>
> "A pool of entities that (a) are considered indistinguishable from
> each other for the purposes of the NETWORK, (b) may participate in
> CHANGES, and (c) are located in a specific PLACE."
>
> "Any kind of ENCOUNTER that can change the quantity of one or more
> STUFF in a NETWORK."
>
> "A ENCOUNTER in which one or more SOLID entities is SOLIDLY
> transformed into one or more other ones. [...]"
>
> "A set or series of ENCOUNTERS, often forming a network, which
> PEOPLE have found useful to group together [...]. NETWORKS can have
> SUBNETWORKS."
>
> Do you think BioPAX or SBML would significantly change of we
> switched the definitions?
>
No - I think I was not able to communicate my point. You are of course
perfectly right that BioPAX and SBML semantics are similar. In fact,
this is something we are trying to do. In the (ideal) future I would
like to see large integrated BioPAX graphs with associated smaller SBML
models and they would map one-to-one.
But - here is the difference. BioPAX wants to build large integrated
graphs, where any portion of the network can be queried and used. This
is a use-case taken by most modern pathway databases. BioCyc, Reactome,
NCI .. they all do that in some form. In SBML every SBML file is a
closed container in itself and is supposed to make sense if you only
used the whole thing. If you can extract a subgraph that can still be
simulated that's great, but not all subgraphs are guaranteed to make
sense. In BioPAX, on the contrary, if you obey some rules you can
reasonably get any subgraph, and it still makes sense. This is a primary
requirement for integration. This is where SBML and BioPAX differs, and
this is where representing particulars in BioPAX becomes a dubious proposal.
> The idea often floated that the COMBINE formats are (or should be)
> non-overlapping makes sense in some cases. For example, graphical
> notation as such is probably orthogonal to everything else (in a
> sense). But BioPAX and SBML are conceptually too close to each other.
> I don't think it is possible to find consensus on a dividing line
> between BioPAX and SBML, let alone enforce such a line, given the
> decentralized development of extensions and packages.
>
I think I just laid out a line.
> It makes more sense to me to go the other way round: acknowledge
> that there is not only an overlap, but a close relationship between
> BioPAX and SBML and define it, i.e. specify a common core of BioPAX
> and SBML (maybe COMBINE Core, or COCORE). I mean, currently, we are
> saying that we have the species glucose in the cytosol which is
> annotated by a MIRIAM qualifier that says that this is the same as the
> physical entity glucose in the cytosol. Why don't we just say this is
> glucose in the cytosol once? Or consider complexes: should we use the
> qualifier "has part" to point to a component, or should we use the
> qualifier "is" pointing to a complex which then has a component, or
> both? Why not have one unified way.
>
Let me repeat, I am all for better mapping, convergence and
coordination. But that's not the discussion we started. We are
discussing why it is (or not) a good idea to include particular (or
existential) data in BioPAX.
> By the way, why don't we build a COMBINE infrastructure? How about
> having two mailing lists, one for all people and one for all COMBINE
> editors, team members and advisers? I would be happy to open two
> Google groups, if there is interest.
>
I, personally, eventually would like to converge towards that as well,
but again - that's a different discussion, isn't it?
Best,
ED
Nicolas Le Novere
The idea often floated that the COMBINE formats are (or should be)
non-overlapping makes sense in some cases. For example, graphical
notation as such is probably orthogonal to everything else (in a
sense). But BioPAX and SBML are conceptually too close to each other.
I don't think it is possible to find consensus on a dividing line
between BioPAX and SBML, let alone enforce such a line, given the
decentralized development of extensions and packages.
BioPAX and SBML are entirely orthogonal. BioPAX is a format to describe biological pathways, i.e. what we think is a biological reality. SBML is a language to encode models, i.e. a formalisation of a problem, valid in a certain context and built for a certain purpose. The fact that some users encode biological pathways in SBML does not make SBML a format meant to encode biological pathways. And of course, you have the limit situation where an hypothetical pathway is built as model to discuss about a pathway. But when it comes to SBML, the main differences with BioPAX are: 1) SBML's "reaction" are processes, not only biochemical reactions 2) SBML's species are pools. Those pools are not necessarily of molecules. They can be pools of cells or organisms. 3) Many SBML models are made of descriptions that have no counterpart in BioPAX. See http://www.ebi.ac.uk/biomodels-main/BIOMD0000000020 http://www.ebi.ac.uk/biomodels-main/BIOMD0000000127 http://www.ebi.ac.uk/biomodels-main/MODEL1008060002 Those models do not have pools of chemicals or descriptions of biochemical reactions.
By the way, why don't we build a COMBINE infrastructure? How about
having two mailing lists, one for all people and one for all COMBINE
editors, team members and advisers? I would be happy to open two
Google groups, if there is interest.
This is on the way :-)
Cheers
Oliver Ruebenacker
On Mon, Jul 11, 2011 at 3:11 PM, Emek Demir <de...@cbio.mskcc.org> wrote:
>> Where would the boundary between BioPAX and SBML be, if there is
>> one? BioPAX already has a few quantities, i.e. equilibrium constant
>> and free energy, with parameters like temperature, pH and pMg. And, of
>> course, stoichiometric coefficients.
>
> Stoichiometric coefficients are completely universal - because of the law of
> mass conservation. Eq. constants and free energy are not unless they are
> specified as formulas where effecting concentrations are variables.
Mass conservation does not always fix stoichiometry. Think of a
reaction where oxygen is on one side and carbon monoxide and carbon
dioxide on the other. I have seen BioPAX data, where the stoichiometry
was essentially an average over different possibilities. If I remember
correctly, it was the number of ATP molecules produced during
respiration in the mitochondria, and the stoichiometry was some odd
fraction. I am not sure this is considered valid use of BioPAX, but on
the other hand, why else are stoichiometric coefficients encoded as
floating point?
But my point was more like this: if stoichiometry is universal, it
falls into the domain of BioPAX. If SBML was truly non-overlapping,
shouldn't that mean that stoichiometry was outside of its domain? And
what does that mean for constraint-based modeling?
>> A concentration is usually scenario-specific. But some
>> concentrations are stable enough to be considered universal.
>
> Example? Stable enough under which context? Can you represent those contexts
> in BioPAX with existing mechanisms? Or do we need to introduce a finer/more
> detailed context representation?
I thought to have heard that some electrolytes are so tightly
controlled that their concentrations can be considered constant. I
might be wrong, though.
>> Where you
>> would draw the line depends on the precision in which you are
>> thinking. A range of possible (or healthy or wild type) concentrations
>> would be a universal statement. What about a 'typical' concentration,
>> would that be universal or specific?
>
> An average, specified together with a variance under some distribution model
> would be a universal -yes..
I think then we are actually in agreement. I thought first, you were
against any sort of concentrations.
>> The line between universal and
>> specific is necessarily blurred, since the universal is the set of all
>> possible particulars
>
> "All swans are white" is a universal statement and it something very
> different than its existential counterpart "I observed a white swan". To
> relate one to the other you need some philosophical structure - you can't
> simply call it a set.This is something often stated in logic theories, so
> from that perspective you are right. But in terms philosophy of science,
> this is an empiricist view most recently argued by logical positivists and
> is considered passé. Popper demolished it in The Logic of Scientific
> Discovery. More accepted modern epistemological views (such as those by
> Popper, Kuhn or their successors) clearly separate a universal from an
> existential. Some of the conflicts that I see between OWL's own semantics
> and science stems from this difference.
It would be fun to discuss the philosophy behind this, but I think
it would be off-topic. But for the record, I think even Popper is not
universally agreed upon (e.g. Martin Gardner seems to disagree), and
Kuhn is actually quite controversial (personally, I think the idea
that Newtonian mechanics is not superior to Aristotelian mechanics is
a view that just does not lead us anywhere).
But what should be the deciding factor is not so much what
philosophers think about it, but what the accepted protocol in the
scientific community is. And there, the particular always serves as a
proxy for the universal. When we ask, what is the value of something,
the answer will always be some sort of weighted average over values
produced by experiments or theory, or, most often, combination of both
(since most values are not measured directly).
>> and can only be learned through the particulars.
>> That's why BioPAX talks about experimental evidence.
>
> Umm.. yes. But BioPAX do not store particulars.. it just points to the
> instances that was used to derive this universal statement. It is there to
> be used purely to determine how much you are going to trust that statement.
> But what you are suggesting is something categorically different.
As long as statements entirely non-quantitative, and can only be
true or false, then the only thing evidence provides is confidence. As
soon as you ave quantitative statements, then you will also want to
know how precise they are.
It boils down to this question: are statements that are universal,
quantitative and approximate within the domain of BioPAX or not?
If they map one-to-one, they are equivalent, aren't they?
> But - here is the difference. BioPAX wants to build large integrated graphs,
> where any portion of the network can be queried and used. This is a use-case
> taken by most modern pathway databases. BioCyc, Reactome, NCI .. they all do
> that in some form. In SBML every SBML file is a closed container in itself
> and is supposed to make sense if you only used the whole thing. If you can
> extract a subgraph that can still be simulated that's great, but not all
> subgraphs are guaranteed to make sense. In BioPAX, on the contrary, if you
> obey some rules you can reasonably get any subgraph, and it still makes
> sense. This is a primary requirement for integration. This is where SBML and
> BioPAX differs, and this is where representing particulars in BioPAX becomes
> a dubious proposal.
If an SBML model is realistic, it should be possible to take parts
of it and those parts should still represent a realistic model of
whatever they describe. And if I take BioPAX data and derive any
conclusion from them, that conclusion depends on what was included in
the data and what was not.
>> The idea often floated that the COMBINE formats are (or should be)
>> non-overlapping makes sense in some cases. For example, graphical
>> notation as such is probably orthogonal to everything else (in a
>> sense). But BioPAX and SBML are conceptually too close to each other.
>> I don't think it is possible to find consensus on a dividing line
>> between BioPAX and SBML, let alone enforce such a line, given the
>> decentralized development of extensions and packages.
>>
> I think I just laid out a line.
The question is whether you can find a consensus :)
>> It makes more sense to me to go the other way round: acknowledge
>> that there is not only an overlap, but a close relationship between
>> BioPAX and SBML and define it, i.e. specify a common core of BioPAX
>> and SBML (maybe COMBINE Core, or COCORE). I mean, currently, we are
>> saying that we have the species glucose in the cytosol which is
>> annotated by a MIRIAM qualifier that says that this is the same as the
>> physical entity glucose in the cytosol. Why don't we just say this is
>> glucose in the cytosol once? Or consider complexes: should we use the
>> qualifier "has part" to point to a component, or should we use the
>> qualifier "is" pointing to a complex which then has a component, or
>> both? Why not have one unified way.
>>
> Let me repeat, I am all for better mapping, convergence and coordination.
> But that's not the discussion we started. We are discussing why it is (or
> not) a good idea to include particular (or existential) data in BioPAX.
Maybe it is a different discussion, but they seem related over the
question of whether BioPAX and SBML are (or should be) overlapping.
>> By the way, why don't we build a COMBINE infrastructure? How about
>> having two mailing lists, one for all people and one for all COMBINE
>> editors, team members and advisers? I would be happy to open two
>> Google groups, if there is interest.
>>
> I, personally, eventually would like to converge towards that as well, but
> again - that's a different discussion, isn't it?
Perhaps it is a different discussion, but they are related.
Oliver Ruebenacker
On Mon, Jul 11, 2011 at 6:13 PM, Nicolas Le Novere <n.len...@gmail.com> wrote:
> On 11 July 2011 19:17, Oliver Ruebenacker <cur...@gmail.com> wrote:
>>
>> The idea often floated that the COMBINE formats are (or should be)
>> non-overlapping makes sense in some cases. For example, graphical
>> notation as such is probably orthogonal to everything else (in a
>> sense). But BioPAX and SBML are conceptually too close to each other.
>> I don't think it is possible to find consensus on a dividing line
>> between BioPAX and SBML, let alone enforce such a line, given the
>> decentralized development of extensions and packages.
>
> BioPAX and SBML are entirely orthogonal.
Orthogonal (i.e. non-overlapping) would mean that any information
that can be expressed in BioPAX can not be expressed in SBML and vice
versa. It would mean that there would be no such thing as a mapping or
a conversion form one to the other, not even an imperfect mapping or a
lossy conversion, but nothing at all; the mere idea would be patently
absurd. No database would even consider exporting the same data to
both BioPAX and SBML, because it would be obviously impossible.
> BioPAX is a format to describe
> biological pathways, i.e. what we think is a biological reality. SBML is a
> language to encode models, i.e. a formalisation of a problem, valid in a
> certain context and built for a certain purpose.
I suppose you can build a model in SBML that has no relationship with
reality and it is still somehow useful to you. But typically, people
try to build realistic models, i.e. models that represent what we
think is a biological reality. Otherwise, it would be impossible to
annotate them or enter them into a curated database.
> The fact that some users
> encode biological pathways in SBML does not make SBML a format meant to
> encode biological pathways. And of course, you have the limit situation
> where an hypothetical pathway is built as model to discuss about a pathway.
Yes, exactly. If that is not an overlap, I wonder what would be?
> But when it comes to SBML, the main differences with BioPAX are:
>
> 1) SBML's "reaction" are processes, not only biochemical reactions
In BioPAX, a conversion is defined as "an interaction in which one
or more physical entities is physically transformed into one or more
other ones".
So how is SBML's reaction different?
> 2) SBML's species are pools. Those pools are not necessarily of molecules.
> They can be pools of cells or organisms.
In BioPAX, a physical entity is defined as "a pool of entities,
where each entity has a physical structure."
Certainly, that includes cells and organisms.
> 3) Many SBML models are made of descriptions that have no counterpart in
> BioPAX. See
> http://www.ebi.ac.uk/biomodels-main/BIOMD0000000020
> http://www.ebi.ac.uk/biomodels-main/BIOMD0000000127
As far as I can see, electric charges and fields are physical
entities (in BioPAX).
Obviously, you can't do the math in BioPAX. But that's not the
point. We all agree that BioPAX and SBML are not equivalent. But to be
orthogonal, you would have to show that no aspect or part of any SBML
model has any counterpart in BioPAX.
> http://www.ebi.ac.uk/biomodels-main/MODEL1008060002
Cool, a model of zombie attack. Zombies are physical entities,
aren't they? Yeah, BioPAX can do zombie attacks!
I suppose if you had chosen a model of demon possession, I would be
in trouble. Although, on second thought - demons are made of fire, and
fire is a physical entity, right?
> Those models do not have pools of chemicals or descriptions of biochemical
> reactions.
>> By the way, why don't we build a COMBINE infrastructure? How about
>> having two mailing lists, one for all people and one for all COMBINE
>> editors, team members and advisers? I would be happy to open two
>> Google groups, if there is interest.
>
> This is on the way :-)
Sounds cool! But why wait, why not have mailing lists right now?
Nicolas Le Novere
>> BioPAX and SBML are entirely orthogonal.
>
> Orthogonal (i.e. non-overlapping) would mean that any information
> that can be expressed in BioPAX can not be expressed in SBML and vice
> versa. It would mean that there would be no such thing as a mapping or
> a conversion form one to the other, not even an imperfect mapping or a
> lossy conversion, but nothing at all;
> the mere idea would be patently
> absurd. No database would even consider exporting the same data to
> both BioPAX and SBML, because it would be obviously impossible.
> Sounds cool! But why wait, why not have mailing lists right now?
This will be done before COMBINE 2011.
--
Nicolas LE NOVERE, Computational Systems Neurobiology, EMBL-EBI, WTGC,
Hinxton CB101SD UK, Mob:+447833147074, Tel:+441223494521 Fax:468,
Skype:n.lenovere, AIM:nlenovere, twitter:@lenovere
http://www.ebi.ac.uk/~lenov/, http://www.ebi.ac.uk/compneur/
Oliver Ruebenacker
On Wed, Jul 13, 2011 at 12:19 PM, Nicolas Le Novere
<n.len...@gmail.com> wrote:
> On 13/07/11 17:09, Oliver Ruebenacker wrote:
>
>>> BioPAX and SBML are entirely orthogonal.
>>
>> Orthogonal (i.e. non-overlapping) would mean that any information
>> that can be expressed in BioPAX can not be expressed in SBML and vice
>> versa. It would mean that there would be no such thing as a mapping or
>> a conversion form one to the other, not even an imperfect mapping or a
>> lossy conversion, but nothing at all;
>
> Absolutely true. The conversion between BioPAX and SBML rely on associated
> tools such as MIRIAM URIs and SBO. Without external information you cannot
> really produce a meaningful conversion. And even with that information, you
> obtain pathological SBML or BioPAX, and the round-tripping is near
> impossible (I would even get rid of the "near")
>
>> the mere idea would be patently
>> absurd. No database would even consider exporting the same data to
>> both BioPAX and SBML, because it would be obviously impossible.
>
> No. The database exports either BioPAX or SBML. The two exports are
> different, and both represent a different vision of the underlying database
> content.
How about the following piece of information:
"There is a process that turns three entities from one identified
pool into five entities of another identified pool."
You can express that information both in BioPAX and SBML. To me,
that is overlap.
But at least we seem to agree that there is even more overlap
between BioPAX and the combination of SBML and annotations.
>> Sounds cool! But why wait, why not have mailing lists right now?
>
> It is on the way. This is not so easy. We do not want to create yet another
> mailing-list where people would need to subscribe, resulting in weakly
> overlapping venn diagrams.
The only alternative to an opt-in list that comes to my mind would
be an opt-out list, which may not be according to every one's taste.
> And we want the list belong to co.mbine.org, to
> shield it against moving around from one systems to another, forcing people
> to re-register (I missed one year of BioPAX discussions because I was not
> aware of the move away from biopax-discuss out of biopax.org).
I don't think you can prevent in any way the possibility that one
day you may want to move. You could, of course, transfer all the
addresses to the new list.
> This will be done before COMBINE 2011.
Looking forward.