david nickerson cellml workshop 2013

David NickersonCellML Workshop 2013

www.sbi.uni-rostock.de

Reproducible simulation experiments with SED-ML

Dagmar Waltemath13.03.2012

Nested Tasks Proposal forSED-ML

Frank T. Bergmann

COMBINE 2012 – Simulation Algorithm, Experiments and Results – Toronto, CA

www.sbi.uni-rostock.de

The necessity for reproducible science

4

Fig.: Example simulation results (Le Novère, Neuroinformatics (2010))

5

Similarly for complex models!

× time× patient× drug dose× …

www.sbi.uni-rostock.de 6

SED-ML Motivation

Simulationtool

models

Biologicalpublicationrepository

?Simulation result

www.sbi.uni-rostock.de

SED-ML Motivation

“[..] in Biomodels database the model BIOMD0000000139 and BIOMD0000000140 are two different models and they are supposed to show different results. Unfortunately simulating them in Copasi gives same result for both the models. [..] “ (arvin mer on sbml-discuss)

Fig.: running model files (COPASI simulation tool)

www.sbi.uni-rostock.de

SED-ML Motivation

“[..] in Biomodels database the model BIOMD0000000139 and BIOMD0000000140 are two different models and they are supposed to show different results. Unfortunately simulating them in Copasi gives same result for both the models. [..] “ (arvin mer on sbml-discuss)

Fig.: running model files (COPASI simulation tool)

www.sbi.uni-rostock.de 9

SED-ML Motivation

Simulation

Simulation results (SBW Workbench)

BIOMD0000000139 , BIOMD0000000140

models

www.sbi.uni-rostock.de

SED-ML Level 1 Version 1

10

Levels:major revisions containing substantial changes

Versions: minor revisions containing corrections and refinements

Editorial board: coordinates SED-ML development (elected by sed-ml-discuss members)

SED-ML Level 1 Version 1:• multiple models • multiple simulation setups• time course simulations only• only explicit model entities can be addressed (XPath)

11

www.sbi.uni-rostock.de

Example: What happens if I just simulate?

12

First attempt to run the model, measuring the spiking rate v over time

load SBML into the simulation tool COPASI use parametrisation as given in the SBML

file define output variables (v) run the time course

1 ms (standard) 100ms 1000ms

www.sbi.uni-rostock.de

Example: What happens if I just simulate?

13

Fig: reference publication Fig.: COPASI simulation, duration: 140ms, step size: 0.14

Second attempt to run the model, adjusting simulation step size and duration

www.sbi.uni-rostock.de

Example: What happens if I just simulate?

14

Fig.: reference publication Fig.: COPASI, adjusted parameter values (a=0.02, b=0.2 c=-55, d=4)

Third attempt to run the model, updating initial model parameters

15

16

17

SED-ML L1V2• Two main additions plus some tidying-up.• Algorithm class proposal.• Nested tasks proposal.

www.sbi.uni-rostock.de

Main building blocks – SED-ML L1V1

18

Figure: SED-ML structure (Waltemath et al., 2011)

www.sbi.uni-rostock.de

SED-ML – What does it look like?

19

KiSAO identifier only

20

Algorithm class proposal• Currently no mechanism to parameterize a particular

simulation algorithm.• Proposal: extend algorithm class to support

parameterization.• Adds a list of algorithm parameters which allows the

author to encode values for the algorithm parameters defined in KiSAO.

<algorithm kisaoID="KISAO:0000032"><listOfAlgorithmParameters>

<algorithmParameter kisaoID="KISAO:0000211" value="23"/></listOfAlgorithmParameters>

</algorithm>

21

Nested tasks proposal

The Current State

22

ModelSimulation

Task

Data Generators

Reports

UniformTimeCourseSimulation

The Current State

• Carry out multiple time course simulations• Collect results from these simulations• Combine results from these simulations• Report / Graph the results

What it does not cover

Parameter Scans

Motivation

• Of course there are more simulations to be captured. But the extensions ought to happen in an orthogonal way, so that we can cover more of what people want to do, by adding additional: – Simulation Classes– Task Classes– DataGenerator Classes– Output Classes

Motivation

• Expand SED-ML, in a natural, compatible way, that allows software tools to carry out interconnected simulation tasks.

Proposal of a new Task Class: The RepeatedTask, that allows multiple simulation tasks to be invoked while changing model variables.

The Proposal

28

ModelSimulation

Task

Data Generators

Reports

• UniformTimeCourseSimulation• oneStep• steadyState

• Task• RepeatedTask

Repeated TaskRepeated Task

resetModel : Booleanrange : IdRef

[0..*]

Range

id : SId

UniformRange

start : doubleend : doublenumberOfPoints: int

VectorRange

value : double[1..*]

FunctionalRange

function: MathML

Repeated TaskRepeated Task

resetModel : Booleanrange : IdRef

[0..*]

Range

id : SId

UniformRange

start : doubleend : doublenumberOfPoints: int

VectorRange

value : double[1..*]

FunctionalRange

function: MathML

SetValue

target : XpathlistOfVariables : Variable[0..*]math : MathML

[0..*]

Repeated TaskRepeated Task

resetModel : Booleanrange : IdRef

SubTask

task : IdRef[0..*]

[0..*]

Range

id : SId

UniformRange

start : doubleend : doublenumberOfPoints: int

VectorRange

value : double[1..*]

FunctionalRange

function: MathML

SetValue

target : XpathlistOfVariables : Variable[0..*]math : MathML

[0..*]

Features

• Carry out simulations, • continue with other simulations• Modify model variables during simulation runs

Parameter Scan

Pulse Experiments

Repeated Stochastic Traces

Time Course Parameter Scan

Place For Orthogonal Extensions…

• Parameterizable Simulation Classes• Storing the last state of a task as a new model• Introduction of new Variables (for accessing

implied variables)• Advanced Post-processing in Data Generators• Access of external data

Available Online

http://sysbioapps.dyndns.org/SED-ML_Web_Tools

Available Online

http://sysbioapps.dyndns.org/SED-ML_Web_Tools

Thank You!

• More Information here:

– http://co.mbine.org/standards/sed-ml/proposal – http://sed-ml.org – http://sysbioapps.dyndns.org – http://frank-fbergmann.blogspot.com

www.sbi.uni-rostock.de

How to contribute to SED-ML

1. Have a look at the current SED-ML Specification document on http://sed-ml.org

2. Try out some of the existing examples http://sed-ml.org and http://sourceforge.net/projects/libsedml

3. Identify what is missing for you to encode your simulation experimental setups - What can you not express?

4. Submit a feature request & post it on the list feature request tracker: http://sourceforge.net/projects/sed-ml mailing list: sed-ml-discuss@lists.sourceforge.net

5. … submit a proposal with example files and prototypeproposal tracker: http://sourceforge.net/projects/sed-ml

41