XML for Model Specification:
Introduction and Workshop
XML for Model Specification:
An Introduction and Workshop
An Introduction to XML in the Neurosciences
Sharon Crook, Arizona State University
An Introduction to NeuroML
Fred Howell, University of Edinburgh
NeuroML for Model Specification in ChannelDB and GENESIS
Dave Beeman, University of Colorado, Boulder
MorphML: An XML Application for Neuronal Morphology Data
Sharon Crook, Arizona State University
Building 3-D Network Models with neuroConstruct
Padraig Gleeson, University College London
Discussion: Current Issues and Future Development
Introduction to XML in the Neurosciences
What is an eXtensible Markup Language (XML) application?
 Portable format for computer documents.
 Data are surrounded by text descriptions called tags.
 Due to the self-describing representation, programs can
parse the data easily.
 Tags are ordinary text and should be clear, concise, and
make sense to humans.
 Language elements that provide the structure make up an
XML schema.
 Each language element may also be equivalent to an object
class.
<!-Segment: mainDend2, ID: 2-->
<segment>
<id>2</id>
<proximal>4</proximal>
<distal>5</distal>
<parent>0</parent>
</segment>
Additional Potential Benefits of XML Schema:
 Validate documents/data.
 Generate instructions for creating database tables for data
element storage and access.
 Easily generate data structures and code for reading and
writing valid XML documents.
 Facilitates communication and collaboration!!!
Potential Liabilities of XML:
 May not be clear, concise, easy to read.
 Most of the advantages of XML can be accomplished with
good discourse, good design, and good documentation.
 Extremely verbose so performance will suffer.
 Private data accessed more easily.
Additional Benefits of XML:

Neuroinformatics infrastructure
(NeuroML Schema and NeuroML Development Kit, MorphML,
BrainML, CellML, SBML)

Commercial/free development software available


Schema development, validation, documentation and more
Altova XMLSpy
http://www.altova.com
XML schema to Java object classes and more
Java Architecture for XML Binding (JAXB)
http://java.sun.com/xml/jaxb
Relevant XML Applications:
 BrainML (http://www.brainml.org)
Laboratory of Neuroinformatics, Weill Medical College of Cornell University
Examples: time series data, spike trains, experimental protocols, recording sites,
bibliographic citations, taxonomy, vital statistics of subject, training statistics,
some attributes of neurons for inheritance
 SBML (http://www.sbml.org)
Systems Biology Markup Language for modeling biochemical reaction networks
Examples: metabolic networks, cell-signaling pathways, regulatory networks
 CellML (http://www.cellml.org)
Bioengineering Institute, University of Auckland
Examples: models of cellular and subcellular processes such as calcium
dynamics, metabolic pathways, signal transduction
 MathML (http://www.w3.org/Math)
W3C World Wide Web Consortium
Examples: mathematical notation with structure and content; serve, receive, and
process mathematics on the web
An introduction to NeuroML
Fred Howell
Adaptive and neural computation
Informatics
University of Edinburgh
fwh@inf.ed.ac.uk
Overview

What are the problems?

Object models, data binding and NeuroML

The next steps?
Scratch pad
●
Ideas for new slides.
Collins et al, J Biol Chem 280:7 2005
Orig version by Ding Fan
Models we’d like to build…
Aim

Move model specifications from programs to a
declarative XML format.
Why XML?

Language independent way to store complex structured
information.

Huge industry momentum.


Not a programming language – so encourages declarative
specifications.
Possible to transform from one format to another –
whereas programs have to be recoded by hand
Why not XML?

Cumbersome to edit by hand

Large text files, need to be compressed

Harder to parse than ad hoc text formats

Not suitable for binary data
Scripts
Parameter
search
Simulation
engine
Results +
visualisations
Custom
extensions
of
simulator
Declarative model
spec (in XML)
Simulation
engine
Results +
visualisations
How would one publish a model?

Put XML model spec on your website, + links to code to
run it.



Plus links back to any experimental data used to derive
parameters / validate results.
See Robert Gentleman's campaign for “reproducible
research”
(and also ModelDB)
Why is this hard?

Lots of levels of scale and detail of models (from protein
interactions to large scale networks of neurons)

Different simulators have different and changing
capabilities – which creates a moving target for attempts
to build any standards
Union or intersection?

Should a model exchange format restrict itself to
a standard subset of possible models, or cope
with any possible model?
What is “NeuroML”?

A way to map from object trees to XML

... with a java development kit

... and some suggestions for sample schemas

channel, cell and network levels

Emphasis on making it easy to define any object model
and serialise it

... create generic tools which work with any object model

... and encourage developers to agree on common
object models where it makes sense
Other XML Languages

SBML : a standard for intracellular pathway models

CellML

MathML
Practicalities

“I'm writing a simulator and I'd like to get the models into NeuroML –
what do I do?”

(1) Separate out the declarative aspects of the model spec

(2) Serialise the model into XML, using the NeuroML
development kit (in Java) or your own code

(3) If any other developers are creating similar models, see if you
can agree on a common set of classes to describe the models by
hand
The NeuroML development kit



A “data binding” kit
Start with class definitions
Utilities to read / write model definition as readable XML
Data binding


A technique for serialising data in objects as XML
Your program can read in an XML document using:


Object o = XMLIn(“file.xml”);
And write one using:


Object o = new MyComplexStructure();
XMLOut(o,”file.xml”);
The XML tags can correspond to fields in the class
<neuroml class=”MyComplexStructure”>
<position>1000</position>
<sequence>ACGGTTCAG</sequence>
<pubmedID>4321652</pubmedID>
</neuroml>
class MyComplexStructure {
int position;
String sequence;
int pubmedID;
}
Example network definition:
package neuroml.model.network;
import neuroml.core.*;
public class Network extends Element {
/** A network has a set of elements - can be populations or individual cells
*/
public Set elements = new Set("ElementRef");
/** A network also defines a set of projections between elements
*/
public Set projections = new Set("Projection");
}
Example class definition:
public class Grid3DStructure extends PopulationStructure {
public int xsize=1;
public int ysize=1;
public int zsize=1;
}
... and so on for all the classes / parameters of your model.
Uses a restricted subset of Java as schema definition language:
int, double, String
Set, Ref, List
classes and inheritance
namespaces
Do code modules / embedded scripts have a
place in NeuroML?
Useful for quickly coding loops for running
simulations, ad hoc connectivity... but perhaps having
any code in the model spec defeats the object?
State of play
simulators adopting own XML formats for serialising
model descriptions
common standards working where the domain is
stable (SBML, MorphML)
The next steps?


How much standardisation is useful?

Just XML in any format?

XML with uniform mapping from classes to <tags>?

A set of rigid standards for compartmental neurons, channels,
receptors, networks, ...?
What features are needed from a development
kit?

C++, python, java?
NeuroML for Model Specification
in ChannelDB and GENESIS
Dave Beeman
University of Colorado, Boulder
WAM-BAMM*05
The Problem:
One neuronal model --> Many implementations
EXAMPLE: Hodgkin-Huxley K channel
Equations with parameter values describe the model.
Simulator scripts tell the simulator how to implement it.
Differences in simulator design
--> NEURON and GENESIS scripts look very different
--> Very difficult to convert a script to one for a
different simulator
The Solution:
Establish a standard format for a declarative representation,
NOT a simulator-dependent procedural representation.
Hodgkin-Huxley K Channel Model
Possible Representations
●
Represent the equations in a form that can be parsed into Java
●
Store tabulated values of rate variables
●
Use parameterized form (A + BV) / (C + D exp((E + V)/F))
The ChannelDB Solution
(http:/www.modelersworkspace.org/channeldb/ChannelDB.html)
XML representation of a Java Hodgkin-Huxley object with
attributes for Gmax, and a set of gates and their exponents
●
Gate objects have attribute telling if it depends on voltage
or concentration, and objects for the forward and backward
rate parameters
●
NeuroML development parser (http://www.neuroml.org)
converts between XML representation and Java objects
●
Use simple Java string manipulation commands to produce
a simulation script from information in the fields of the
DBChannel object
●
Prototype database and interface creates commented
GENESIS scripts from stored XML channel descriptions
●
NeuroML representation of the Hodgkin-Huxley K
channel
<neuroml class="DBChannel" description="Hodgkin-Huxley squid K channel"
author="Dave Beeman" keywords="Hodgkin-Huxley potassium squid delayed rectifier"
uniqueID="10262778758662F22@dogstar.colorado.edu"
notes="An implemention of the GENESIS K_squid_hh channel" Erest="-0.07V">
<channels>
<channel name="K_squid_hh" class="HHChannel" permeantSpecie="K"
Erev="0.09V" Gmax="360.0S/m^2" ivlaw="ohmic">
<gates>
<gate name="X" class="HHVGate" timeUnit="sec" voltageUnit="V"
vmin="-0.1" vmax="0.05"
instantCalculation="false" useState="false" power="4">
<forwardRate class="ParameterizedHHRate" A="-600.0"
B="-10000.0" C="-1.0" D="1.0" E="0.060" F="-0.01"/>
<backwardRate class="ParameterizedHHRate" A="125.0" B="0.0
C="0.0" D="1.0" E="0.07" F="-0.08"/>
</gate>
</gates>
<log author="Dave Beeman" date="Jul 9, 2002 11:11:15 PM"
literatureReference="A.L. Hodgkin and A.F. Huxley,
J. Physiol. (Lond) 117, pp 500-544 (1952)">
<logEntries>
</logEntries>
</log>
</channel>
</channels>
</neuroml>
Some classes defined for ChannelDB
DBChannel: Wrapper class that is used to contain any channel model that is stored in ChannelDB, along
with some descriptive information.
HHChannel: Class used for all the Hodgkin-Huxley type channels in the database.
HHVGate: Used as a member of the gates set of a HHChannel. It contains forward and backward rate
objects that depend on voltage, as well as some additional fields to describe the gate.
HHCGate: An ionic concentration-dependent gate, analogous to the voltage-dependent HHVGate. It
provides an additional field for a reference to the object that provides the source of the ionic concentration.
HHRate: The superclass for the specialized forms for the rate variables.
ParameterizedHHRate: A subclass of HHRate that expresses rate variables in a parameterized form
typical of many Hodgkin-Huxley type rate equations, "rate = (A + BV) / (C + D exp((E + V)/F))"
EquationHHRate: A subclass of HHRate that expresses the rate variables as equations.
TabulatedHHRate: A subclass of HHRate that allows a gate's forwardRate or backwardRate to be
specified by a table at equally spaced voltage (or concentration) points.
ConcenPool: Describes a single shell model for a concentration pool, with a buildup of concentration
proportional to an incoming current and a time constant for decay. The object providing the source of
concentration to a HHCGate is typically formed from this class. The source of currents is provided by a set
of objects of class CurrentSource.
CurrentSource: Used by ionic concentration pools to provide information about the object that provides an
ionic current.
Unfinished Business and Open Questions
Extend NeuroML to provide representations for more detailed multi-shell models of
calcium diffusion
Implement a more sophisticated representation of literature references than the simple
string that is currently used in the NeuroML software. (We have proposed a schema for
the Modeler's Workspace based on BibTeX.)
Software to convert ChannelDB descriptions to NEURON and other simulators
Implement the HHCVGate, a two-dimensional gate depending on both voltage and
concentration. (Note that the Traub Ca-dependent K channel model uses a form that
can be expressed as a product of a HHVGate and a HHCGate.)
Implement Borg-Graham or Lytton-Sejnowski temperature-dependent channel models
with the NeuroML ThermodynamicHHVGate.
Is there a better way for a concentration-dependent channel model to reference the
models that provide the source of ionic currents and concentrations?
How much standardization should there be for the format and the names of the
independent variables and parameters in equation representations?
Unbundling GENESIS
GENESIS 3 Core – Based on MOOSE
The Messaging Object Oriented Simulation Environment
a reimplementation of GENESIS base code in C++
by U. S. Bhalla, NCBS, Bangalore
Provides:
Improved Messaging between GENESIS objects
●
●
Faster, smaller, cleaner implementation
●
Portable to MS Windows and non-UNIX platforms
●
Improved equation solvers
●
Allows multiple parsers and interfaces
GENESIS 3 will add:
Graphical interface
● XML representation of models
● Backwards compatibility with GENESIS 2
● Tutorials and educational applications
●
Planned GENESIS 3 Interfaces
An XML Application for
Neuronal Morphology Data
http://www.morphml.org
Sharon Crook
Arizona State University
Department of Mathematics and Statistics
School of Life Sciences
WAM-BAMM*05
MorphML XMLSpy Documentation
WAM-BAMM*05
MorphML XMLSpy Documentation
WAM-BAMM*05
MorphML: A Simple Example from neuroConstruct
<?xml version="1.0" encoding="UTF-8"?>
<n:morphml xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
xmlns:n="http://morphml.org/morphml/schema/1.0.0"
xsi:schemaLocation="http://morphml.org/morphml/schema/1.0.0
http://math.la.asu.edu/~crook/morphml/MorphML.xsd">
<n:name>SimpleCell</n:name>
<n:notes>A Simple cell for testing purposes</n:notes>
<n:lengthUnits>Micrometers</n:lengthUnits>
<!--Converting cell: SimpleCell-->
<n:points>
<!-Start point of segment: Soma, ID: 0-->
<n:point>
<n:id>0</n:id>
<n:x>0.0</n:x>
<n:y>0.0</n:y>
<n:z>0.0</n:z>
<n:diameter>16</n:diameter>
</n:point>
<!-End point of segment: Soma, ID: 0-->
<n:point>
<n:id>1</n:id>
<n:x>0.0</n:x>
<n:y>0.0</n:y>
<n:z>0</n:z>
<n:diameter>16</n:diameter>
</n:point>
<!-Start point of segment: mainDend1, ID: 1-->
<n:point>
<n:id>2</n:id>
<n:x>0.0</n:x>
<n:y>0.0</n:y>
<n:z>0.0</n:z>
<n:diameter>2</n:diameter>
</n:point>
WAM-BAMM*05
MorphML: A Simple Example from neuroConstruct
<!-End point of segment: mainDend1, ID: 1-->
<n:point>
<n:id>3</n:id>
<n:x>-10.0</n:x>
<n:y>-30.0</n:y>
<n:z>0</n:z>
<n:diameter>2</n:diameter>
</n:point>
<!-Start point of segment: mainDend2, ID: 2-->
<n:point>
<n:id>4</n:id>
<n:x>0.0</n:x>
<n:y>0.0</n:y>
<n:z>0.0</n:z>
<n:diameter>2</n:diameter>
</n:point>
<!-End point of segment: mainDend2, ID: 2-->
<n:point>
<n:id>5</n:id>
<n:x>10.0</n:x>
<n:y>-30.0</n:y>
<n:z>0</n:z>
<n:diameter>2</n:diameter>
</n:point>
<!-Start point of segment: mainAxon, ID: 3-->
<n:point>
<n:id>6</n:id>
<n:x>0.0</n:x>
<n:y>0.0</n:y>
<n:z>0.0</n:z>
<n:diameter>2</n:diameter>
</n:point>
WAM-BAMM*05
MorphML: A Simple Example from neuroConstruct
<n:cells>
<n:cell>
<n:name>SimpleCell</n:name>
<!-Segments of the cell
-->
<n:segments>
<!-Segment: Soma, ID: 0-->
<n:segment>
<n:id>0</n:id>
<n:proximal>0</n:proximal>
<n:distal>0</n:distal>
</n:segment>
<!-Segment: mainDend1, ID: 1-->
<n:segment>
<n:id>1</n:id>
<n:proximal>2</n:proximal>
<n:distal>3</n:distal>
<n:parent>0</n:parent>
</n:segment>
<!-Segment: mainDend2, ID: 2-->
<n:segment>
<n:id>2</n:id>
<n:proximal>4</n:proximal>
<n:distal>5</n:distal>
<n:parent>0</n:parent>
</n:segment>
<!-Segment: mainAxon, ID: 3-->
<n:segment>
<n:id>3</n:id>
<n:proximal>6</n:proximal>
<n:distal>7</n:distal>
<n:parent>0</n:parent>
</n:segment>
WAM-BAMM*05
MorphML: A Simple Example from neuroConstruct
<!-Segment: subAxon1, ID: 4-->
<n:segment>
<n:id>4</n:id>
<n:proximal>8</n:proximal>
<n:distal>9</n:distal>
<n:parent>3</n:parent>
</n:segment>
<!-Segment: subAxon2, ID: 5-->
<n:segment>
<n:id>5</n:id>
<n:proximal>10</n:proximal>
<n:distal>11</n:distal>
<n:parent>3</n:parent>
</n:segment>
</n:segments>
</n:cell>
</n:cells>
</n:morphml>
WAM-BAMM*05
Virtual Ratbrain (http://www.ratbrain.org)
Laszlo Zaborszky, Peter Varsanyi
Center for Molecular and Behavioral Neuroscience, Rutgers
Fred Howell, Nicola McDonnell
Institute of Adaptive and Neural Computation, University of Edinburgh
•
•
•
Database for peer reviewed 3-D cellular anatomical
data of the rat brain
Visualization and analysis tools including analysis of
dendritic and axonal morphometry
Data stored in MorphML format
WAM-BAMM*05
Virtual Ratbrain (http://www.ratbrain.org)
MorphML Viewer
WAM-BAMM*05
Building 3D Network Models with
neuroConstruct
(Summary of main presentation)
Padraig Gleeson
University College London
p.gleeson@ucl.ac.uk
WAM-BAMM*05
31 March 2005
Scope of Application
●
Reuses existing base of models/modellers
●
Adds functionality
–
Graphical interface
–
Checks on morphologies
–
Network building capabilities
–
Storage/replay/analysis of simulations
●
Built with Java: runs on any platform
●
Code produced is native GENESIS/NEURON
Visualization
●
Single Cells can be viewed in 3D
●
Information on morphology
●
Checks on consistency of cell structure
●
Segments can be edited
●
Info on basic electrophysiology
Screenshot: Cell Visualization
Packing in 3D
●
●
Cell Groups are packed in 3D Regions
–
Rectangular Box
–
Spherical
Various Packing Patterns
–
Random
–
Cubic close packed
–
Hexagonal
–
Single positioned
–
Evenly spaced in 1D
Screenshot: Packing in 3D
Simulator Interaction(1)
●
●
Morphology files can be imported from
–
GENESIS (*.p readcell format files)
–
NEURON (most ntscable like files, with create, pt3dadd, connect)
–
Cvapp (SWC format files)
–
MorphML
Imported cells are checked for validity: i.e. errors
which may cause problems on some platforms
–
zero length segments
–
all except root segment have parents
–
unique names, etc.
Simulator Interaction(2)
●
●
Files can currently be exported to:
–
NEURON, for simulation
–
GENESIS, for simulation
–
MorphML, for publishing/use by other simulators
Cell info held in simulator independent format
–
Can be mapped to other/future simulators
Cell Processes
●
●
Generic models of Cell Process
(channels/synapses) can be used in
neuroConstruct
Model separated from experimentally measured
parameters
●
Reuse of tried and tested template files
●
Can be mapped on to any simulator
●
Automatic handling of units
Modularity of Cell Processes (1)
Pre-existing & well tested
XML template of model of
Cell Process, e.g. Double Exp
Synapse, HH Channel
Mapping of templates to
existing simulators
Experimentally determined
parameter set:
gmax, Tau Rise/Decay, etc.
Published model of
Cell Process
(XML file)
Modularity of Cell Processes (2)
neuroConstruct
Parameters coupled with GENESIS/
NEURON template placed on cells
GENESIS/NEURON
source code
Published model of
Cell Process
Parameters with
plotting module
Plots of Cell Process
internals
Screenshot: Cell Processes
Morphology mapping (1)
●
neuroConstruct Concepts
–
Section: unbranched part of axon/dendrite with the same biophysical
properties
–
Segment: Specifies one 3D point along Section, shaped like conical
frustum
–
Section specifies start point, Segments specify 3D points along
Morphology mapping (2)
●
●
Going from GENESIS -> NEURON
–
Reasonably straightforward
–
Compartments in GENESIS mapped to segments in neuroConstruct
Going from NEURON -> GENESIS
–
Non-trivial: mapping conical sections to cylinders
–
Simple mapping: each segment to compartment with equivalent
surface area
NeuroML/MorphML interaction
●
neuroConstruct currently allows:
–
●
Import & export of MorphML morphologies
Future support
–
Greater support for specification of groups/Cell Process locations in
MorphML format
–
Importation of Cell Processes in NeuroML format
–
Export of network structure in NeuroML format
–
Generation of simulation code in NeuroML/NeoSim format
XML for Model Specification:
Introduction and Workshop
XML for Model Specification: Discussion
Wider/Easier Access to NeuroML:
1. Would a CVS server for materials on the website that
allows for multiple contributors/editors be of use?
2. Would a separation of the NeuroML schema into several,
more focused schemas (ex: morphology, channels,
channel distributions, connectivity) be useful for making it
more transparent and easier to use? If so, where do we
define relationships among the language elements of each
separate schema?
3. An XMLSpy HTML version of the specs might help with
documentation.
4. Tight coupling of NeuroML with Java? What about other
languages?
XML for Model Specification: Discussion
Channel Issues:
1. Can we use one XML channel specification for both
NEURON and GENESIS specs?
2. How can we make XML channel documents easy to
use for someone who is not a GENESIS or NEURON
user?
3. Details of channel distributions in current NeuroML
Schema, GENESIS, and NEURON? Can we define a
schema that includes them all?
XML for Model Specification: Discussion
Morphology Issues:
Mapping morphology across formats (MorphML, GENESIS,
NEURON, BBT, SWC).
1. One issue is that GENESIS uses cylindrical
compartments while others can be frustums. You can
map between compartments of equal surface area
which accounts for the membrane parameters but not
the axial resistance.
2. In Neuron, the 3-D points describing a cable/section
and the actual numerical integration points (nseg
points evenly spaced) are different.
3. Segment of length zero for cell body?
4. MorphML->NEURON->MorphML connectivity?
XML for Model Specification: Discussion
Future Development:
1. Relation to other XML applications? BrainML, CellML and
SBML?
2. Some of the inbuilt XML features in Java can be useful and
further schema development should take advantage of
these (ex: XML support in Java J2SE 5.0).
3. What concrete plans do different groups have over the
next year for XML-related developments?
The End