- From: m.cooling at auckland.ac.nz (Mike Cooling)
- Subject: [cellml-discussion] FW: Representing stochastic models in CellML
- Date: Fri, 23 May 2008 18:33:33 +1200
Just a note, for some reason you don't now need the SOCKs proxy from in the
ABI.
>
-----Original Message-----
>
From: cellml-discussion-bounces at cellml.org
>
[mailto:cellml-discussion-bounces at cellml.org] On Behalf Of Randall
>
Britten
>
Sent: Monday, 19 May 2008 4:59 p.m.
>
To: 'CellML Discussion List'
>
Cc: 'Kevin Burrage'
>
Subject: Re: [cellml-discussion] Representing stochastic models in CellML
>
>
Hi
>
>
I'd prefer Skype at that time.
>
>
I've found the Skype conference calls work fairly well, easiest if one
>
person initiates the call, and all others make sure that the initiator
knows
>
their skype-name, and that they want to be called to join in. Also, Skype
>
from within the ABI works well too, though you'll need IT to explain how
to
>
use the SOCKS proxy, or to set it up for you if you have not done it
before.
>
>
Regards,
>
Randall
>
>
> -----Original Message-----
>
> From: cellml-discussion-bounces at cellml.org [mailto:cellml-discussion-
>
> bounces at cellml.org] On Behalf Of Andrew
>
> Sent: Monday, 19 May 2008 4:17 p.m.
>
> To: Kevin Burrage
>
> Cc: cellml-discussion at cellml.org
>
> Subject: Re: [cellml-discussion] Representing stochastic models in
>
> CellML
>
>
>
> Kevin Burrage wrote:
>
>> Andrew
>
>>
>
>> i am at conference early next week and then lectures and meetings
>
>> wednesday and thursday - i am free friday 23 at say 8.30am - but how
>
>> are we going to do it. I am not sure that i can access the grid room
>
>> - can skype work?
>
>> BW
>
> Hi Kevin,
>
>
>
> I am happy with 7:30 PM NZST (8:30 AM for you) on Friday the 23rd.
>
>
>
> I have got Skype working at home (and I have seen it working on the UoA
>
> network for other people although not always); we could try running it
>
> as a conference call and this might allow more people to participate
>
> (but may also be more prone to not working).
>
>
>
> Perhaps the best thing would be for everyone who is interested, and
>
> whether they would be happy with the time, and whether they would
>
> rather
>
> conference call in or (for people in Auckland) be in the same room at
>
> the ABI.
>
>
>
> Best regards,
>
> Andrew
>
>
>
>> Kevin
>
>> -------------------------------------------------
>
>> Kevin Burrage
>
>> Professor of Computational Systems Biology, COMLAB, University of
>
> Oxford
>
>> and Professor of Computational Mathematics, IMB, University of
>
> Queensland
>
>> -------------------------------------------------
>
>>
>
>>
>
>>
>
>> On Wed, 14 May 2008 11:58:03 +1200
>
>> Andrew Miller <ak.miller at auckland.ac.nz> wrote:
>
>> Kevin Burrage wrote:
>
>>> Andrew
>
>>>
>
>>> I have lots of ideas on this - but email not a good way of doing it.
>
>>> Am in oxford at the moment but could skype you if you were
>
> interested.
>
>> Hi Kevin,
>
>>
>
>> Sorry for the delay in getting back to you about this.
>
>>
>
>> There seems to be a lot of interest amongst members of
>
>> the CellML community in Auckland about hearing your views
>
>> on this; one suggestion I heard was to use the BestGRID
>
>> rather than Skype so more people could be involved at this
>
>> end (although I am not sure how convenient it would be for
>
>> you to access the BestGRID facilities in Oxford so the
>
>> timezones would work out).
>
>>
>
>> When would be the most convenient for you? Given that
>
>> there are a number of people who have indicated a desire
>
>> to be involved in these discussion, and also a few that
>
>> are interested in sitting in to listen, I would suggest
>
>> making it at earliest next week, to ensure everyone can
>
>> get their schedules to fit, and to ensure that software is
>
>> working ahead of time. Since you are, I presume, in a
>
>> GMT+1 timezone, and we are in GMT+12, the best time would
>
>> probably be early in the morning for you / night time for
>
>> us, e.g. 7-8 AM Oxford / 6-7 PM NZ, 8-9 AM Oxford / 7-8 PM
>
>> NZ , or 9-10 AM Oxford / 8-9 PM NZ.
>
>>
>
>> Best regards,
>
>> Andrew
>
>>
>
>>> BW
>
>>>
>
>>> Kevin
>
>>> -------------------------------------------------
>
>>> Kevin Burrage
>
>>> Professor of Computational Systems Biology, COMLAB, University of
>
> Oxford
>
>>> and Professor of Computational Mathematics, IMB, University of
>
>>> Queensland
>
>>> -------------------------------------------------
>
>>>
>
>>>
>
>>>
>
>>> On Wed, 23 Apr 2008 17:31:26 +1200
>
>>> Andrew Miller <ak.miller at auckland.ac.nz> wrote:
>
>>> Hi all,
>
>>>
>
>>> I have spent some time looking into how we might be able
>
>>> to represent stochastic models in CellML. This is
>
>>> something that would be useful to ensure we have properly
>
>>> contemplated for CellML 1.2. I have pasted in the notes I
>
>>> wrote on this below.
>
>>>
>
>>> Please let me know if you have any suggestions, comments,
>
>>> or criticisms of the below document. At some point, this
>
>>> will obviously need to be transformed into a more robust
>
>>> proposal, but for now, I just want to make sure we keep
>
>>> the option open to use the CellML 1.2 core to represent
>
>>> stochastic models.
>
>>>
>
>>> Best regards,
>
>>> Andrew
>
>>>
>
>>> -----
>
>>>
>
>>> Overall goal:
>
>>> Describe a framework which can be used in CellML 1.2 to
>
>>> represent a
>
>>> range of
>
>>> different systems which require stochastic differential
>
>>> equations to
>
>>> describe.
>
>>>
>
>>> Constraints:
>
>>> Do not want to describe the procedure for solving the
>
>>> model in core
>
>>> CellML,
>
>>> only the underlying mathematical / statistical model.
>
>>> Want to express the model in such a way that the
>
>>> procedure is computable
>
>>> from the model.
>
>>> Want there to be only one interpretation of the model.
>
>>> Want the representation to be abstract enough that it is
>
>>> meaningful for a
>
>>> number of different fields, and not just chemical
>
>>> equations in a well
>
>>> stirred vessel.
>
>>> Want the representation to work naturally when mixed
>
>>> with systems of
>
>>> ordinary
>
>>> differential equations.
>
>>>
>
>>> Use cases:
>
>>> Chemical reactions under the Chemical Master Equation
>
>>> model of Gillespie:
>
>>> We need to split these into separate species. This is
>
>>> a Poisson process,
>
>>> so there are simple ways to represent it.
>
>>>
>
>>> It is more efficient to represent models using Weiner
>
>>> processes when
>
>>> this
>
>>> there are large enough numbers of molecules to justify
>
>>> this but ODEs are
>
>>> not being used.
>
>>>
>
>>> However, Poisson and Weiner processes are both Levy
>
>>> Processes, that is,
>
>>> they have stationary independent increments, are zero
>
>>> at time zero, and
>
>>> are cadlag. This is not necessarily a good thing
>
>>> because some things we
>
>>> want to model might have memory of past events or a
>
>>> time dependence.
>
>>>
>
>>> How we can represent this in CellML:
>
>>> For the continuous case, integral equations for the
>
>>> increment in terms of
>
>>> built in processes like Weiner and Poisson processes (I
>
>>> don't believe
>
>>> there
>
>>> is a clean way to represent increments in MathML).
>
>>>
>
>>> Implementations will need to identify these and work out
>
>>> the
>
>>> distribution of
>
>>> the time until the next event (good implementations
>
>>> might be able to
>
>>> perform
>
>>> symbolic algebra to work this out, but most
>
>>> implementations would probably
>
>>> just recognise common cases like Poisson distributions
>
>>> with arbitrary
>
>>> parameters, and deal with expressions involving a Weiner
>
>>> process by
>
>>> sampling
>
>>> from the increment distribution in each time step),
>
>>> which could be put
>
>>> into a
>
>>> slightly generalised Gibson-Bruck type of framework
>
>>> where we store the
>
>>> time of
>
>>> the next event.
>
>>>
>
>>> None of this helps for non stationary independent
>
>>> processes, however.
>
>>>
>
>>> How could this be related to the standards:
>
>>> It has been proposed that CellML 1.2 have a core
>
>>> specification which
>
>>> describes
>
>>> the basic way of representing the mathematical structure
>
>>> in very general
>
>>> terms, and secondary specifications be used to narrow
>
>>> CellML down to
>
>>> specific
>
>>> subsets which can be implemented in their entirety by a
>
>>> actual software
>
>>> packages.
>
>>>
>
>>> Core CellML 1.2 should be general enough to represent
>
>>> concepts like
>
>>> probability distributions (MathML allows new operators
>
>>> to be defined,
>
>>> so we
>
>>> could create ones for our base types of stochastic
>
>>> process). The ODE IV
>
>>> secondary specification would not allow stochastic
>
>>> models, while we would
>
>>> have another alternative secondary specification which
>
>>> extended the ODE IV
>
>>> one to allow certain limited types of stochastic model
>
>>> (limited to
>
>>> types we
>
>>> know how to solve).
>
>>>
>
>>> In terms of interaction with the typing system, in a
>
>>> stochastic system we
>
>>> have both reals and real-valued random variables. Once
>
>>> we have one random
>
>>> variable in our system, this will propagate to
>
>>> everything else which
>
>>> is affected by it, so most of the model will technically
>
>>> be a random
>
>>> variable.
>
>>> However, we want to be able to easily mix stochastic
>
>>> models with
>
>>> existing ODE
>
>>> IV models to create hybrid models, so we don't really
>
>>> want the required
>
>>> datatype to change just to connect up a random variable
>
>>> to a non-random
>
>>> variable. For this reason, I don't think it is
>
>>> worthwhile to consider a
>
>>> random variable of a certain type a different datatype,
>
>>> and instead we
>
>>> would
>
>>> require tools to deduce this information if they require
>
>>> it.
>
>>>
>
> _______________________________________________
>
> cellml-discussion mailing list
>
> cellml-discussion at cellml.org
>
> http://www.cellml.org/mailman/listinfo/cellml-discussion
>
>
_______________________________________________
>
cellml-discussion mailing list
>
cellml-discussion at cellml.org
>
http://www.cellml.org/mailman/listinfo/cellml-discussion
>
- [cellml-discussion] FW: Representing stochastic models in CellML, Mike Cooling, 05/23/2008
Archive powered by MHonArc 2.6.18.