p548_lecture7_CompuCell_intro

Download Report

Transcript p548_lecture7_CompuCell_intro

Introduction to CompuCell3D
Outline:
1. What is CompuCell3D?
2. Why use CompuCell3D?
3. Demo simulations
4. Glazier-Graner-Hogeweg (GGH) model – an overview
5. CompuCell3D architecture and terminology
6. XML 101
7. Building first CompuCell3D simulation
8. Visualization package – CompuCell Player
9. Python scripting inside CompuCell3D
What Is CompuCell3D?
1. CompuCell3D is a modeling environment used to build, test, run and visualize
GGH-based simulations
2. CompuCell3D has built-in scripting language (Python) that allows users to quite
easily write extension modules that are essential for building sophisticated
biological models.
3. CompuCell3D thus is NOT a specialized software
4. Running CompuCell3D simulations DOES NOT recompilation
5. CompuCell3D model is described using CompuCell3D XML syntax and in the
case of using Python language , a Python script(s)
6. CompuCell3D platform is distributed with a GUI front end – CompuCell Player
or simply Player. The Player provides 2- and 3-D visualization capabilities.
7. Models developed by one CompuCell3D user can be “replayed” by another user
regardless the operating system/hardware on which CompuCell is installed.
8. CompuCell3D is a cross platform application with Windows port being currently
developed (due end of July 2007).
Why Use CompuCell3D? What Are the Alternatives?
1. CompuCell3D allows users to set up and run their simmulations within minutes,
maybe hours. A typical development of a specialized GGH code takes orders of
magnitudes longer time.
2. CompuCell3D simulations DO NOT need to be recompiled. If you want to
change parameters (in XML or Python scripts) or logic (in Python scripts) you
just make the changes and re-run the simulation. With hand-compiled
simulations there is much more to do. Recompilation of every simulation is also
error prone and often limits users to those who have significant programming
background.
3. CompuCell3D is actively developed , maintained and supported. On
www.compucell3d.org website users can download manuals, tutorials and
developer documentation. CompuCell3D has approx. 10 releases each year.
4. CompuCell3D has many users around the world. This makes it easier to
collaborate or exchange modules and results saving time spent on developing
new model.
5. The Biocomplexity Institute organizes training workshops and mentorship
programs. Those are great opportunities to visit Bloomington and learn
biological modeling using CompuCell3D. For more info see
www.compucell3d.org
Demo Simulations
The GGH Model – an Overview
x 20
•Energy minimization formalism
- extended by Graner and Glazier, 1992
•DAH: Contact energy depending on cell types (differentiated cells)
E   J ( ( x )), ( ( x ')) (1    ( x ), ( x ') ) 
x, x'
s ( s  S ) 2  v (v  V ) 2 
Echem  Ehapt  ...
•Metropolis algorithm: probability of configuration change
Brief Explanation of Equation Symbols
(x) –denotes id of the cell occupying
position x. All pixels pointed by arrow have
same cell id , thus they belong to the same
cell
((x)) denotes cell type of cell with id (x). In the
picture above blue and yellow cells have different
cell types and different cell id. Arrows mark
different cell types
Notice that in your model you may (will) have many cells of the same type but with
different id. For example in a simple cellsorting simulation there will be many cells of
type “Condensing” and many cells with type “NonCondensinig”
CompuCell3D Architecture
Object oriented implementation in C++ and Python
Visualization, Steering,
User Interface
Python Interpreter
Biologo Code Generator
Kernel
Runs Metropolis Algorithm
Plugins
Calculate change
in energy
PDE Solvers
Lattice monitoring
Typical “Run-Time” Architecture of CompuCell
CompuCellPlayer
CompuCell can be run in a
variety of ways:
•Through the Player with or
without Python interpreter
Python
•As a Python script
CompuCell3D
Kernel
Plugins
•As a stand alone
computational kernel+plugins
CompuCell3D terminology
1. Spin-copy attempt is an event where program randomly picks a lattice site in an
attempt to copy its spin to a neighboring lattice site.
2. Monte Carlo Step (MCS) consists of series spin-copy attempts. Usually the
number of spin copy-attempts in single MCS is equal to the number of lattice
sites, but this is can be customized
3. CompuCell3D Plugin is a software module that either calculates an energy term
in a Hamiltonian or implements action in response to spin copy (lattice
monitors). Note that not all spin-copy attempts will trigger lattice monitors to run.
4. Steppables are CompuCell3D modules that are run every MCS after all spincopy attempts for a given MCS have been exhausted. Most of Steppables are
implemented in Python. Most customizations of CompuCell3D simulations is
done through Steppables
5. Steppers are modules that are run for those spin-copy attempts that actually
resulted in energy calculation. They are run regardless whether actual spin-copy
occurred or not. For example cell mitosis is implemented in the form of stepper.
6. Fixed Steppers are modules that are run every spin-copy attempt.
CompuCell3D Terminology – Visual Guide
Change pixel
Spin copy “blue” pixel
(newCell) replaces
“yellow” pixel (oldCell)
100x100x1 square lattice = 10000 lattice sites (pixels)
MCS 21
MCS 22
10000 spincopy attempts
MCS 23
10000 spincopy attempts
MCS 24
10000 spincopy attempts
10000 spincopy attempts
Run
Run
Run
Steppables
Steppables
Steppables
Nearest neighbors in 2D and their Euclidian distances from the central pixel
Nearest Neighbor Order
5
4
3
4
5
4
2
1
2
4
3 4 5
1 2 4
1 3
1 2 4
3 4 5
1
2
3
4
Number of nearest
neighbors
4
4
4
8
5
4
Euclidian distance – square
lattice
1
2
2
5
8
Spin copy can take place between any order nearest neighbor (although in practice
we limit ourselves to only few first oders).
<FlipNeighborMaxDistance>1.45</FlipNeighborMaxDistance> 2nd nearest neighbor
Contact energy calculation (see further slides) are also done up to certain order of
nearest neighbors (default is 1)
<Depth>2.1</Depth>
XML 101
XML stands for eXtensible Markup Manguage. It is NOT a programming language.
Its main purpose is to standarize information exchange between different
applications.
XML Example:
<Sentence>
<Text>It is too early to be in class</Text>
<FontType>TimesNewRoman</FontType>
<FontSize>12</FontSize>
<DisplayHint Hint=“AddFrameAround”/>
</Sentence>
CompuCell Related Example
Defining basic properties of the simulation like lattice dimension, number of
Monte Carlo Steps, Temperature and ratio of spin-copy attempts to number
of lattice sites (Flip2DimRatio). <Potts> section has to be included in every
CompuCell3D simulation
<Potts>
<Dimensions x="71" y="36" z="211"/>
<Steps>10</Steps>
<Temperature>2</Temperature>
<Flip2DimRatio>2</Flip2DimRatio>
</Potts>
Defining properties of Volume Energy term – cell target volume and lambda
parameter:
<Plugin Name=“Volume">
<TargetVolume>25</TargetVolume>
<LambdaVolume>2.0</LambdaVolume>
</Plugin>
...
Building Your First CompuCell3D Simulation
All simulation parameters are controlled by the config file. The config file
allows you to only add those features needed for your current simulation,
enabling better use of system resources.
Define Lattice and Simulation Parameters
Cell
< CompuCell3D>
<Potts>
<Dimensions x=“100" y=“100" z=“1"/>
<Steps>10</Steps>
<Temperature>2</Temperature>
<Flip2DimRatio>1</Flip2DimRatio>
</Potts>
…
</CompuCell3D>
Define Cell Types Used in the Simulation
Each CompuCell3D xml file must list all cell types that will used in the simulation
Cell
<Plugin Name="CellType">
<CellType TypeName="Medium" TypeId="0"/>
<CellType TypeName=“Light" TypeId="1"/>
<CellType TypeName=“Dark" ="2"/>
</Plugin>
Notice that Medium is listed with TypeId =0. This is both convention and a
REQUIREMENT in CompuCell3D. Reassigning Medium to a different TypeId may
give undefined results. This limitation will be fixed in one of the next CompuCell3D
releases
Define Energy Terms of the Hamiltonian and Their Parameters
Volume
volume
volumeEnergy(cell)
Surface
Cell
area
surfaceEnergy(cell)
Contact
contactEnergy(
cell1, cell2)
<Plugin Name="Volume">
<TargetVolume>25</TargetVolume>
<LambdaVolume>1.0</LambdaVolume>
</Plugin>
<Plugin Name="Surface">
<TargetSurface>21</TargetSurface>
<LambdaSurface>0.5</LambdaSurface>
</Plugin>
<Plugin Name="Contact">
<Energy Type1="Medium" Type2="Medium">0
</Energy>
<Energy Type1="Light" Type2="Medium">0
</Energy>
<Energy Type1="Dark" Type2="Medium">0.1
</Energy>
<Energy Type1="Light" Type2="Light">0.5
</Energy>
<Energy Type1="Dark" Type2="Dark">3.0
</Energy>
<Energy Type1="Light" Type2="Dark">0.5
</Energy>
</Plugin>
Plugin XML Syntax
E  ...  v (v  V ) 2  ...
<Plugin Name="Volume">
<TargetVolume>25</TargetVolume>
<LambdaVolume>1.0</LambdaVolume>
</Plugin>
E  ...  s ( s  S ) 2  ...
<Plugin Name="Surface">
<TargetSurface>21</TargetSurface>
<LambdaSurface>0.5</LambdaSurface>
</Plugin>
Plugin XML Syntax – Contact Energy
E  ...   J ( ( x )), ( ( x ')) (1    ( x ), ( x ') )  ...
x, x'
<Plugin Name="Contact">
<Energy Type1="Medium" Type2="Medium">0
</Energy>
<Energy Type1="Light" Type2="Medium">0
</Energy>
<Energy Type1="Dark" Type2="Medium">0.1
</Energy>
<Energy Type1="Light" Type2="Light">0.5
</Energy>
<Energy Type1="Dark" Type2="Dark">3.0
</Energy>
<Energy Type1="Light" Type2="Dark">0.5
</Energy>
</Plugin>
1- term ensures that pixels belonging to the same cell do not contribute to contact
energy
Laying Out Cells on the Lattice
Using built-in cell field initializer:
<Steppable Type="BlobInitializer">
<Gap>0</Gap>
<Width>5</Width>
<CellSortInit>yes</CellSortInit>
<Radius>40</Radius>
</Steppable>
This is just an example of cell field initializer. More general ways of cell field
initialization will be discussed later.
Putting It All Together - cellsort_2D.xml
<CompuCell3D>
<Potts>
<Dimensions x="100" y="100" z="1"/>
<Steps>10</Steps>
<Temperature>2</Temperature>
<Flip2DimRatio>1</Flip2DimRatio>
</Potts>
<Plugin Name="CellType">
<CellType TypeName="Medium" TypeId="0"/>
<CellType TypeName=“Light" TypeId="1"/>
<CellType TypeName=“Dark" ="2"/>
</Plugin>
<Plugin Name="Volume">
<TargetVolume>25</TargetVolume>
<LambdaVolume>1.0</LambdaVolume>
</Plugin>
<Plugin Name="Surface">
<TargetSurface>21</TargetSurface>
<LambdaSurface>0.5</LambdaSurface>
</Plugin>
<Plugin Name="Contact">
<Energy Type1="Medium" Type2="Medium">0
</Energy>
<Energy Type1="Light" Type2="Medium">0
</Energy>
<Energy Type1="Dark" Type2="Medium">0.1
</Energy>
<Energy Type1="Light" Type2="Light">0.5
</Energy>
<Energy Type1="Dark" Type2="Dark">3.0
</Energy>
<Energy Type1="Light" Type2="Dark">0.5
</Energy>
</Plugin>
<Steppable Type="BlobInitializer">
<Gap>0</Gap>
<Width>5</Width>
<CellSortInit>yes</CellSortInit>
<Radius>40</Radius>
</Steppable>
</CompuCell3D>
Coding the same simulation in C/C++/Java/Fortran would take you at least 1000 lines
of code…
Putting It All Together - Avoiding Common Errors in XML code
1. First specify Potts section, then list all the plugins and finally list all the
steppables. This is the correct order and if you mix e.g. plugins with steppables
you will get an error. Remember the correct order is
•
Potts
•
Plugins
•
Steppables
2. Remember to match every xml tag with a closing tag
<Plugin>
…
</Plugin>
3. Watch for typos
4. Modify available examples rather than starting from scratch
CompuCellPlayer – the Best
Way To Run Simulations
Steering bar allows users to start or pause the
simulation, zoom in , zoom out, to switch between
2D and 3D visualization, change view modes (cell
field, pressure field , chemical concentration field,
velocity field etc..)
Player can output multiple
views during single
simulation run – Add
Screenshot function
Information bar
Opening a Simulation in the Player
Go to File->Open Simulation ; Click Simulation xml file -> Browse… button
Running Simulation From Command Line
You can simply start the simulation with or without Player straight from command line
Open up console (terminal) and type:
./compucell3d.command –i cellsort_2D.xml (on OSX)
./compucell3d.command –i cellsort_2D.xml (on Linux)
Running CompuCell3D from command line not only convenient, but sometimes (on
clusters) the only option to run the simulation. For more information about command
line options please see “Running CompuCell3D” manual available at
www.compucell3d.org.
Running the Simulation
•After typing the XML file in your favorite editor all you need to do to run the
simulation is to open the XML file in the Player and hit “Play” button.
•Screenshots from the simulations are automatically stored in the directory with
name composed of simulation file name and a time at which simulation was started
•As you can see this setting CompuCell3D simulation was reasonably simple.
•It is quite likely that if you were to code entire simulation in C/C++/Java etc. you
would need much more time.
•We hope that now you understand why using CompuCell3D saves you a lot of time
and allows you to concentrate on biological modeling and not on writing low level
computer code.
•During last year we have improved CompuCell3D performance so that it is on par
with hand-written code. Yet, if you really to have the fastest GGH code in the world
you should write code your own simulation directly in C or even better in assembly
language. Before you do it, make sure you want to spend time rewriting the code
that already exist. We hope you will enjoy CompuCell3D and for more information
please visit www.compucell3d.org.
Capabilities of CompuCellPlayer
•Provides wide range of visualization - cell field plots, concentration plots, vector
field plots in both 2- and 3-D
•Allows to store multiple lattice views in a single run. For example users can store
multiple projections of the cell lattice, concentration fields, various 3D views etc… in
a single run.
•Can be run in GUI and silent mode (i.e. without displaying GUI but still saving
screenshots)
•Is ready to be used on clusters that do not have X-server installed. This feature is
essential for doing “production runs” of your simulations.
•Concentration fields and vector fields initialized from Python level can easily be
displayed in the Player
•Configurable from XML level for those users who prefer typing to clicking
CompuCell Player 101
XML initializers - UniformInitializer
You may initialize simple geometries of cell clusters directly from XML
<Steppable Type=“UniformInitializer">
<Region>
<BoxMin x=“10” y=“10” z=“0”/>
<BoxMax x=“90” y=“90” z=“1”/>
<Types>Condensing,NonCondensing</Types>
<Gap>0</Gap>
<Width>5</Width>
</Region>
</Steppable>
Specify box size and position
Specify cell types – here the box will be filled
with cells whose types are randomly chosen
(either 1 or 2)
Choose cell size and space between cells
<Steppable Type=“UniformInitializer">
<Region>
<BoxMin x=“10” y=“10” z=“0”/>
<BoxMax x=“90” y=“90” z=“1”/>
<Types>Condensing</Types>
<Gap>0</Gap>
<Width>5</Width>
</Region>
</Steppable>
Notice, we have only specified one type (Condensing) thus all the cells are of the same
type
<Steppable Type=“UniformInitializer">
<Region>
<BoxMin x=“10” y=“10” z=“0”/>
<BoxMax x=“90” y=“90” z=“1”/>
<Types>Condensing,NonCondensing</Types>
<Gap>2</Gap>
<Width>5</Width>
</Region>
</Steppable>
Introducing a gap between cells
<Steppable Type="UniformInitializer">
<Region>
<BoxMin x="10" y="10" z="0"/>
<BoxMax x="40" y="40" z="1"/>
<Gap>0</Gap>
<Width>5</Width>
<Types>Condensing,NonCondensing</Types>
</Region>
<Region>
<BoxMin x="50" y="50" z="0"/>
<BoxMax x="80" y="80" z="1"/>
<Gap>0</Gap>
<Width>3</Width>
<Types>Condensing</Types>
</Region>
</Steppable>
Notice, we have defined two regions with different cell sizes and different types
XML initializers - BlobInitializer
<Steppable Type="BlobInitializer">
<Region>
<Radius>30</Radius>
<Center x="40" y="40" z="0"/>
<Gap>0</Gap>
<Width>5</Width>
<Types>Condensing,NonCondensing</Types>
</Region>
<Region>
<Radius>20</Radius>
<Center x="80" y="80" z="0"/>
<Gap>0</Gap>
<Width>3</Width>
<Types>Condensing</Types>
</Region>
</Steppable>
Defining two regions with different cell sizes and different types for BlobInitializer is
very similar to the same task with UniformInitilizer. There are some new XML tags
which differ the two initializers.
Using PIFInitilizer
Use PIFInitializer to create sophisticated initial conditions. PIF file allows you to
compose cells from single pixels or from larger rectangular blocks
The syntax of the PIF file is given below:
Cell_id Cell_type x_low x_high y_low y_high z_low z_high
Example (file: amoebae_2D_workshop.pif):
0 amoeba 10 15 10 15 0 0
This will create rectangular cell with x-coordinates ranging from 10 to 15
(inclusive), y coordinates ranging from 10 to 15 (inclusive) and z coordinates
ranging from 0 to 0 inclusive.
0,0
<Steppable Type="PIFInitializer">
<PIFName>amoebae_2D_workshop.pif</PIFName>
</Steppable>
Let’s add another cell:
Example (file: amoebae_2D_workshop.pif):
0 Amoeba 10 15 10 15 0 0
1 Bacteria 35 40 35 40 0 0
Notice that new cell has different cell_id (1) and different type (Bacterium)
Let’s add pixels and blocks to the two cells
from previous example:
Example (file: amoebae_2D_workshop.pif):
0 Amoeba 10 15 10 15 0 0
1 Bacteria 35 40 35 40 0 0
0 Amoeba 16 16 15 15 0 0
1 Bacteria 35 37 41 45 0 0
To add pixels, start new pif line with existing cell_id (0 or 1 here ) and specify pixels.
This is what happens when you do not reuse
cell_id
Example (file: amoebae_2D_workshop.pif):
0 Amoeba 10 15 10 15 0 0
1 Bacteria 35 40 35 40 0 0
0 Amoeba 16 16 15 15 0 0
2 Bacteria 35 37 41 45 0 0
Introducing new cell_id (2) creates new cell.
PIF files allow users to specify arbitrarily complex cell shapes and cell arrangements.
The drawback is, that typing PIF file is quite tedious task and , not recommended.
Typically PIF files are created using scripts.
In the future release of CompuCell3D users will be able to draw on the screen cells or
regions filled with cells using GUI tools. Such graphical initialization tools will greatly
simplify the process of setting up new simulations. This project has high priority on our
TO DO list.
PIFDumper - yet another way to create initial condition
PIFDumper is typically used to output cell lattice every predefined number of MCS. It is
useful because, you may start with rectangular cells, “round them up” by running
CompuCell3D , output cell lattice using PIF dumper and reload newly created PIF file
using PIFInitializer.
<Steppable Type="PIFDumper“ Frequency=“100”>
<PIFName>amoebae</PIFName>
</Steppable>
Above syntax tells CompuCell3D to store cell lattice as a PIF file every 100 MCS.
The files will be named amoebae.100.pif , amoebae.200.pif etc…
To reload file , say amoebae.100.pif use already familiar syntax:
<Steppable Type="PIFInitializer">
<PIFName>amoebae.100.pif</PIFName>
</Steppable>
Practical way of guessing contact energy hierarchy
Basic facts:
•Cells that have high contact energies between themselves, when they come together
they increase overall energy of the system.
•Cells that have low contact energies between themselves, when they come together
they decrease overall energy of the system.
•Those two rules are helpful when determining contact energy hierarchy. Simply cells of
one type like to be surrounded by those cells with which the contact energy is the
lowest.
•And vice versa, if you want to make two cells not to touch each other, make sure that
contact energy between them is high.
Examples of different contact energy hierarchies
Cell sorting simulation where cells of both
type like to be surrounded by medium. That
is contact energy between Condensing and
Medium as well as between
NonCondensing and Medium is very low
JCM=JNM<JNN<JCC<JNC
Examples of different contact energy hierarchies
Cell sorting simulation where cells of both
type do not like to be surrounded by
medium and cells of homotypic cells do not
like each other
JNM<<JNN=JCC<JCM=JNM
Chemotaxis
Basic facts
•Chemotaxis is defined as cell motion induced by a presence (gradient) of a chemical.
•In GGH formalism chemotaxis is implemented as a spin copy bias which depends on
chemical gradient.
•Chemotaxis was first introduced to GGH formalism by Paulien Hogeweg from University
of Utrecht, Netherlands
•In CompuCell3D Chemotaxis plugin provides wide range of options to support different
modes of chemotaxis.
•Chemotaxis plugin requires the presence of at least one concentration field. The fields
can be inserted into CompuCell3D simulation by means PDE solvers or can be created,
initialized and managed explicitly from the Python level
Chemotaxis Term – Most Basic Form
Echem   (c( xdestination )  c( xsource))
If concentration at the spin-copy destination pixel (c(xdestination)) is higher than
concentration at the spin-copy source (c(xsource)) AND  is positive then E is negative
and such spin copy will be accepted. The cell chemotacts up the concentration gradient
C(x)
Lower concentration
Higher concentration
x
Chemorepulsion can be obtained by making  negative
Alternative Formulas For Chemotaxis Energy
Echem
Echem
 c( xdestination )
c( xsource) 

  

 a  c( xdestination ) a  c( xsource ) 
 c( xdestination )
c( xsource) 

  

 a  c( xdestination )  1 a  c( xsource)  1 
Chemotaxis - XML Examples
Echem   (c( xdestination )  c( xsource))
<Plugin Name="Chemotaxis">
<ChemicalField Source="FlexibleDiffusionSolverFE" Name="FGF">
<ChemotaxisByType Type="Amoeba" Lambda="300"/>
<ChemotaxisByType Type="Bacteria" Lambda="200"/>
</ChemicalField>
<ChemicalField Source="FlexibleDiffusionSolverFE" Name="FGF4">
<ChemotaxisByType Type="Amoeba" Lambda=“-300"/>
</ChemicalField>
</Plugin>
Notice , that different cell types may have different chemotactic properties. For more
than 1 chemical fields the change of chemotaxis energy expression is given below:
Echem 
   (c ( x
i  field
i
i
destination
)  ci ( xsource ))
Chemotaxis - XML Examples continued
<Plugin Name="Chemotaxis">
<ChemicalField Source="FlexibleDiffusionSolverFE" Name="FGF">
<ChemotaxisByType Type="Amoeba" Lambda="300"/>
<ChemotaxisByType Type="Bacteria" Lambda="200"/>
</ChemicalField>
<ChemicalField Source="FlexibleDiffusionSolverFE" Name="FGF4">
<ChemotaxisByType Type="Amoeba" Lambda=“-300“ SaturationCoef=“2.0”/>
</ChemicalField>
</Plugin>
Echem
 c( xdestination )
c( xsource) 

  

 a  c( xdestination ) a  c( xsource ) 
Chemotaxis - XML Examples continued
Echem   (c( xdestination )  c( xsource))
<Plugin Name="Chemotaxis">
<ChemicalField Source="FlexibleDiffusionSolverFE" Name="FGF">
<ChemotaxisByType Type="Amoeba" Lambda="300"/>
<ChemotaxisByType Type="Bacteria" Lambda="200"/>
</ChemicalField>
<ChemicalField Source="FlexibleDiffusionSolverFE" Name="FGF4">
<ChemotaxisByType Type="Amoeba" Lambda=“-300“ SaturationLinearCoef=“2.0”/>
</ChemicalField>
</Plugin>
Echem
 c( xdestination )
c( xsource) 

  

 a  c( xdestination )  1 a  c( xsource)  1 
PDE Solvers
•CompuCell3D has built-in diffusion , reaction diffusion and advection diffusion PDE
solvers. Those are, probably most frequently used solver in GGH modeling.
•CompuCell3D uses explicit (unstable but fast) method to solve the PDE. Constantly
changing boundary conditions practically rule out more robust, but slow implicit solvers.
•Because of instability users should make sure that their PDE parameters do not
produce wrong results (which could manifest themselves as “rough” concentration
profiles, “insane” concentration values, NaN’s - Not A Number etc…). Future release of
CompuCell3D will provide tools to detect potential PDE instabilities.
•Additional solvers can be implemented directly in C++ or using BioLogo. BioLogo is
especially attractive because it takes as an input human readable PDE description and
generates fast C++ code.
•Typically a concentration from the PDE solver is read by other CompuCell3D modules
to adjust cell properties. Currently the best way of dealing with this is through Python
interface.
Flexible Diffusion Solver
<Steppable Type="FlexibleDiffusionSolverFE">
Define diffusion field
<DiffusionField>
Define diffusion parameters
<DiffusionData>
<FieldName>FGF</FieldName>
<DiffusionConstant>0.010</DiffusionConstant>
<DecayConstant>0.000</DecayConstant>
<ConcentrationFileName>diffusion_2D.pulse.txt</ConcentrationFileName>
</DiffusionData>
</DiffusionField>
</Steppable>
Read-in initial condition
Initial Condition File Format:
x y z concentration
Example:
27 27 0 2000.0
45 45 0 0.0 …
Two-pulse initial condition
Initial condition (diffusion_2D.pulse.txt):
5 5 0 1000.0
27 27 0 2000.0
<Steppable Type="FlexibleDiffusionSolverFE">
<DiffusionField>
You may specify diffusion regions
<DiffusionData>
<FieldName>FGF</FieldName>
<DiffusionConstant>0.010</DiffusionConstant>
<DecayConstant>0.000</DecayConstant>
<DoNotDiffuseTo>Medium</DoNotDiffuseTo>
<ConcentrationFileName>diffusion_2D.pulse.txt</ConcentrationFileName>
</DiffusionData>
</DiffusionField>
</Steppable>
FGF will diffuse inside big cell and will not go to Medium
<Steppable Type="FlexibleDiffusionSolverFE">
<DiffusionField>
<DiffusionData>
<FieldName>FGF</FieldName>
<DiffusionConstant>0.010</DiffusionConstant>
<DecayConstant>0.000</DecayConstant>
<DoNotDiffuseTo>Wall</DoNotDiffuseTo>
<ConcentrationFileName>diffusion_2D_wall.pulse.txt</ConcentrationFileName>
</DiffusionData>
</DiffusionField>
</Steppable>
FGF will not diffuse to the Wall
<Steppable Type="FlexibleDiffusionSolverFE">
<DiffusionField>
<DiffusionData>
<FieldName>FGF</FieldName>
<DiffusionConstant>0.010</DiffusionConstant>
<DecayConstant>0.000</DecayConstant>
<!--DoNotDiffuseTo>Wall</DoNotDiffuseTo-->
<ConcentrationFileName>diffusion_2D_wall.pulse.txt</ConcentrationFileName>
</DiffusionData>
</DiffusionField>
</Steppable>
Now FGF diffuses everywhere
PDE Solver Caller Plugin
By default PDE solver is called once per MCS. You may call it more often, say 3 times
per MCS by including PDESolverCaller plugin:
<Plugin Name="PDESolverCaller">
<CallPDE PDESolverName="FlexibleDiffusionSolverFE" ExtraTimesPerMC=“2"/>
</Plugin>
Notice, that you may include multiple CallPDE tags to call different PDESolvers with
different frequencies.
You typically use this plugin to avoid numerical instabilities when working with large
diffusion constants
Secretion
CompuCell3D offers several modes for including secretion in your simulations. Let’s look
at concrete examples:`
<Steppable Type="FlexibleDiffusionSolverFE">
<DiffusionField>
<DiffusionData>
<FieldName>FGF</FieldName>
<DiffusionConstant>0.000</DiffusionConstant>
<DecayConstant>0.000</DecayConstant>
</DiffusionData>
<SecretionData>
<Secretion Type="Amoeba">20</Secretion>
</SecretionData>
</DiffusionField>
</Steppable>
We turned diffusion off and have cells of type
Amoba secrete FGF. Secretion takes place at every
pixel belonging to Amoeba cells. At each MCS we
increase the value of the concentration at those
pixels by 20 units.
<Steppable Type="FlexibleDiffusionSolverFE">
<DiffusionField>
<DiffusionData>
<FieldName>FGF</FieldName>
<DiffusionConstant>0.000</DiffusionConstant>
<DecayConstant>0.000</DecayConstant>
</DiffusionData>
<SecretionData>
<SecretionOnContact Type=“Amoeba" SecreteOnContactWith=“Medium">20.1</SecretionOnContact>
</SecretionData>
</DiffusionField>
</Steppable>
Secretion will take place in those pixels
belonging to Amoeba cells that have contact
with Medium
<Steppable Type="FlexibleDiffusionSolverFE">
<DiffusionField>
<DiffusionData>
<FieldName>FGF</FieldName>
<DiffusionConstant>0.000</DiffusionConstant>
<DecayConstant>0.000</DecayConstant>
</DiffusionData>
<SecretionData>
<SecretionOnContact Type="Medium" SecreteOnContactWith="Amoeba">20.1</SecretionOnContact>
</SecretionData>
</DiffusionField>
</Steppable>
Secretion will take place in those pixels
belonging to Medium cells that have contact
with Amoeba
<Steppable Type="FlexibleDiffusionSolverFE">
<DiffusionField>
<DiffusionData>
<FieldName>FGF</FieldName>
<DiffusionConstant>0.000</DiffusionConstant>
<DecayConstant>0.000</DecayConstant>
</DiffusionData>
<SecretionData>
<SecretionOnContact Type="Medium" SecreteOnContactWith="Amoeba">20.1</SecretionOnContact>
<SecretionOnContact Type="Bacteria“ SecreteOnContactWith="Bacteria">10.1</SecretionOnContact>
<SecretionOnContact Type="Bacteria" SecreteOnContactWith="Medium">5.1</SecretionOnContact>
</SecretionData>
</DiffusionField>
</Steppable>
1.Secretion will take place in those pixels
belonging to Medium cells that have contact
with Amoeba.
2.There will be secretion in pixels of Bacteria
cells that have contact with medium.
3.Secretion will also take place in those pixels
of bacteria cells that have contact with other
bacteria cells
More Flexible Specification of Surface and Volume Constraints
<Plugin Name="VolumeFlex">
<VolumeEnergyParameters CellType=“Amoeba" TargetVolume=“150" LambdaVolume="10"/>
<VolumeEnergyParameters CellType=“Bacteria" TargetVolume=“10" LambdaVolume=“50"/>
</Plugin>
You may specify different volume and surface constraints for different cell types. This
can be done entirely at the XML level.
E  V  (v  V ) 2
Type dependent quantities
E  S  ( s  S ) 2
<Plugin Name=“SurfaceFlex">
<SurfaceEnergyParameters CellType=“Amoeba" TargetSurface=“60" LambdaSurface="10"/>
<SurfaceEnergyParameters CellType=“Bacteria" TargetSurface=“12" LambdaSurface=“20"/>
</Plugin>
Even More Flexible Specification of Surface and Volume Constraints
<Plugin Name="VolumeLocalFlex“/>
E  V  (v  V ) 2
<Plugin Name=“SurfaceLocalFlex“/>
E  S  ( s  S ) 2
Notice that all the parameters are local to a cell. Each cell might have different target
volume (target surface) and different  volume (surface). You will need to use Python to
initialize or manipulate those parameters while simulation is running. There is currently
no way to do it from XML level. I am not sure it would be practical either.
Tracking Cell Neighbors
Sometimes in your simulation you need to have access to a current list of cell neighbor.
CompuCell3D makes this task easy:
<Plugin Name=“NeighborTracker“/>
Inserting this statement in the plugins section of the XML will ensure that at any given
time the list of cell neighbors will be accessible to the user. You can access such a list
either using C++ or Python. In addition to storing neighbor list , a common surface area
of a cell with its neighbors is stored.
Tracking Center of Mass of Each Cell
Including
<Plugin Name=“CenterOfMass“/>
statement in your XML code (remember to put it in the correct place) will enable cell
centroid tracking:
x C CM 
 xi
yCM 
i  pixel
y
i  pixel
z C CM 
i
z
i  pixel
i
To get a center of mass of cell you will need to divide centroids by the cell volume:
xCM
x C CM

V
yCM
y C CM

V
zCM
z C CM

V
Imposing Directed Motion of Cells
One can impose artificial spin flip bias that would have an effect of moving cell in the
direction OPPOSITE to Lambda vector specified below. The magnitude of the lambda
vector determines the “amount” of spin copy bias.
<Plugin Name="ExternalPotential">
<Lambda x="-0.5" y="0.0" z="0.0"/>
</Plugin>
Eexternal _ potential   ( x destination  x source )
E will be negative (favoring spin copy)

Connectivity Plugin
Connectivity plugin ensures that 2D cells are not fragmented and are simply
connected. It decreases probability of certain spin flips which are can break
connectedness of a cell. Users can specify energy penalty that will be incured if the
spin copy is to break connectedness of the cell.:
Syntax:
<Plugin Name=“Connectivity”>
<Penalty>100000</Penalty>
</Plugin>
Cell sorting simulation with and without connectivity plugin
Length Constraint Plugin
Length constraint plugin is used to force cells to keep preferred length along cell’s
longest axis (we assume that cells have elliptical shape):
<Plugin Name=“LengthConstraint”>
<LengthEnergyParameters TargetLength=“15” LambdaLength=“2.0”/>
</Plugin>
The LambdaLength and TargetLength play similar role to LambdaVolume and
TargetVolume from Volume Plugin.
IMPORTANT: Length Constraint Plugin has to be used together with connectivity plugin
or else cells might become fragmented. The applicability of the LengthConstraint and
Connectivity Plugins is limited to 2D simulations.
For more information see
“Cell elongation is key to in silico replication of in vitro vasculogenesis and subsequent
remodeling” by Roeland M.H. Merks et al Developmental Biology 289 (2006) 44– 54
Length constraint plugin at work
scripting inside
CompuCell3D
XML gives you the ability to change simulation parameters using human-readable
syntax but does not allow users to implement more complex cell behaviors,
sophisticated cell type transition rules, inter-cell signaling or connecting to
intracellular models
Python scripting capabilities in CompuCell3D allow users to accomplish
abovementioned tasks (and much more) and are the reasons why CompuCell3D
is called simulation environment, not simulation application.
Python scripting capabilities allow users to use rich set of Python modules and
third party libraries and provide level flexibility comparable with packages such as
Matlab or Mathematica
Python Scripting Prerequisites
•To make full use of Python scripting users should be familiar with Python
programming language. They do not need to be experts though.
•CompuCell3D comes with template and example codes that make Python scripting
inside CompuCell3D quite easy even for beginners.
•Python scripting in CompuCell3D typically requires users to develop a class that
implements required logic. If you are unfamiliar with concept of class , think of it as a
type that has several data members (such as floats, integers, other classes) and set
of functions that operate on those internal members but can also take external
arguments. Class is a generalization of “C” structure or “Pascal” record.
•Fortunately CompuCell3D comes with plenty of examples that users can adapt to
serve their needs. This does not require thorough programming knowledge. If you
are unfamiliar with Python scripting, reading (and doing) “CompuCell3D Python
Scripting Tutorials” should quickly get you up-to-speed.
Where Do You Begin?
•You will still need XML file describing which plugins and/or steppables
CompuCell3D is to use. In the future releases we might change this requirement but
for now it appeared as the best way of ensuring proper module initialization
•In addition to XML file you will need to write Python script that implements main
CompuCell3D logic i.e. reads XML file, initializes modules and executes calls in a
loop Metropolis algorithm. This file will also call set up and initialize your modules
written in Python. CompuCell3D comes with many examples of such files so in fact
preparing one is reduced to minor modification of existing one.
•Once you have XML and Python scripts ready you opent them up in the Player and
start running simulaitons
What Can You Do in Python?
You may implement any CompuCell3D module using Python – energy functions,
lattice monitors, steppers, steppables, fixed steppers.
You need to remember that Python is an interpreted language and thus executes
orders of magnitudes slower than for example C++. This means that although you
can easily develop energy functions (remember, they are the most frequently called
modules in CompuCell3D) in Python, you will probably want to avoid using them with
your “production runs”. In this it makes sense to implement those functions in C++ ,
which is not too difficult and we provide comprehensive Developers documentation.
Since lattice monitors are called less frequently than energy functions, the
performance degradation due to lattice monitor being implemented in Python is
much smaller. That same is true for steppers, fixed steppers and steppables.
Notice that CompuCell3D kernel that is called from Python, as well as other core
CompuCell3D modules called from Python run at native speeds, as they are
implemented in C++ with only their API exposed to Python. Therefore if you run
CompuCell3D through Python script but decide not to implement new Python
modules, your speed of run will be essentially identical as if you ran CompuCell3D
using just XML file.
What are the advantages of using Python inside CompuCell3D
Rapid development – no compilation is necessary. Write or modify your script and
run
Portability – script developed by you on one machine (e.g. Mac) is ready to use
under linux
Model integration - you can quite easily implement hooks to subcellular models. We
have been able to set up CompuCell3D simulation that was using SBW network
intracell simulators within few minutes. T
Rich set of external libraries – you may tap into rich Python library repositories that
exist for essentially any task
Agile development – developing and refining your model is an iterative process.
Working at the compiled language stage will force you to spend significant portion of
your time waiting for the program to compile. With Python you eliminate this step
thus increase productivity. Users should first prototype their models in Python and
once they are ready to be used for production runs, rewrite the once causing
significant slowdown in C++.
Your first CompuCell3D Python script. Make sure you have your copy of Python
Scripting Tutorials
Begin with template code (the file will be called CellInfoPrinter.py)
import sys
from os import environ
import string
sys.path.append(environ["PYTHON_MODULE_PATH"])
import CompuCellSetup
import CompuCell sim,simthread = CompuCellSetup.getCoreSimulationObjects()
#Create extra player fields here or add attributes or plugin:
#PUT YOUR CODE HERE
CompuCellSetup.initializeSimulationObjects(sim,simthread)
steppableRegistry=CompuCellSetup.getSteppableRegistry()
#Add Python steppables here
#PUT YOUR CODE HERE
CompuCellSetup.mainLoop(sim,simthread,steppableRegistry)
Opening a simulation with Python script in the Player
Go to File->Open Simulation ; Choose xml file. Click Python script
“Browse…” button to select python script. Do not forget to check “ Run
Python script” check box!
Hopefully you are not scared. All the code presented above is a template code with
one place holder to register your newly developed steppable. We will first teach you
how to develop a steppable because steppables are most likely to be developed in
Python anyway.
Let’s take a look at the module that prints cell id, cell type and cell volume for every
cell in the simulation. Iterating over all cells is probably most frequently used task in
steppables:
class InfoPrinterSteppable(SteppablePy):
def __init__(self,_simulator,_frequency=10):
SteppablePy.__init__(self,_frequency)
self.simulator=_simulator
self.inventory=self.simulator.getPotts().getCellInventory()
self.cellList=CellList(self.inventory)
def start(self):
print "This function is called once before simulation"
def step(self,mcs):
print "This function is called every 10 MCS“
for cell in self.cellList:
print "CELL ID=",cell.id, " CELL TYPE=",cell.type," volume=",cell.volume
Python Steppable
Each Python Steppable should have three functions:
start()
step(mcs)
finish()
It is OK to leave the implementation of any of above functions empty. An empty function
will be then called.
In addition to this, because Python steppables are implemented as classes they need to
define __init__ function that act as a constructor.
Steppable Template:
class YourPythonSteppable(SteppablePy):
def __init__(self,_simulator,_frequency=10):
#your code here
def start(self):
#your code here
def step(self,mcs):
#your code here
def finish(self):
#your code here
If you are non-programmer it may looks a bit strange, but imagine how much more
would be required to write do the same in C/C++. Much more. Let’s explain the code:
class InfoPrinterSteppable(SteppablePy):
def __init__(self,_simulator,_frequency=10):
SteppablePy.__init__(self,_frequency)
self.simulator=_simulator
self.inventory=self.simulator.getPotts().getCellInventory()
self.cellList=CellList(self.inventory)
First line defines our steppable class. Each class has to have __init__method that is
called when object of this class is created. You can pass any arguments to this
method, but the first argument must be “self”. This is required by Python language.
First line in __init__ method initializes Base class SteppablePy. Do not worry if you do
not understand it. Treat it as a boiler plate code.
Line self.simulator=_simulator stores a pointer or reference to simulator object as a
member variable. This way, later on whenever we decide to reference simulator we
would use self.simmulator instead of getting simulator some other , more complicated
way. Notice, that whenever you access member variable you prepend their name with
keyword “self”
Subsequently we get reference to cell inventoryt (C++ object) and use it to create
iterable cellList (self.cellList=CellList(self.inventory)) Notice, “self” shows up again.
def step(self,mcs):
print "This function is called every 10 MCS“
for cell in self.cellList:
print "CELL ID=",cell.id, " CELL TYPE=",cell.type," volume=",cell.volume
Above function implements core functionality of our steppable. It informs that it is called
every 10 MCS – see how we set frequency parameter in the __init__ function.
The last two lines do actual iteration over each cell in the cell inventory
Notice that it is really easy to do the iteration:
for cell in self.cellList:
Now you can see how storing CellType object as self.cellList comes handy. All we need
to do is to pass iterable cell list (self.cellList) to the “for” loop.
Actual printing is done in line
print "CELL ID=",cell.id, " CELL TYPE=",cell.type," volume=",cell.volume
For each cell in inventory “cell” variable of the for loop will be initialized with different cell
from inventory. All you need to do is to print cell.id, cell.type, and cell.volume. It is pretty
simple.
You need to remember this:
Python distinguishes blocks of codes by their indentation. Therefore
for cell in self.cellList:
print "CELL ID=",cell.id
print " CELL TYPE=“, cell.type
print " volume=",cell.volume
would result in an error because the line print " volume=",cell.volume has different
indentation than other print statement and thus does belong to the “for” loop. Python
will attempt executing this line once after the “for” loop is finished and will return an
error that global object “cell” was not found. It was found because “cell” name is valid
only inside the “for” loop body. Since the last line was not in the body, you get an error.
We are using 3 spaces to indent block of codes, you may choose differently, but need
to be consistent.
Now save the file with the steppable as , say, MySteppables.py . All you need to do is to
provide hooks to your steppable in the main Python script:
steppableRegistry=CompuCellSetup.getSteppableRegistry()
########## Steppable Registration ###########
from MySteppables import InfoPrinterSteppable
infoPrinter= InfoPrinterSteppable(sim)
steppableRegistry.registerSteppable(infoPrinter)
##########End of Steppable Registration ###########
steppableRegistry.init(sim)
Notice that registering steppable requires importing your steppable from the file:
from MySteppables import InfoPrinterSteppable
creating steppable object:
infoPrinter= InfoPrinterSteppable(sim)
registering it with steppable registry:
steppableRegistry.registerSteppable(infoPrinter)
Full Main Script:
import sys
from os import environ
from os import getcwd
import string
sys.path.append(environ["PYTHON_MODULE_PATH"])
sys.path.append(getcwd()+"/examples_PythonTutorial")
import CompuCellSetup
sim,simthread = CompuCellSetup.getCoreSimulationObjects()
#Create extra player fields here or add attributes
CompuCellSetup.initializeSimulationObjects(sim,simthread)
#Add Python steppables here
steppableRegistry=CompuCellSetup.getSteppableRegistry()
from cellsort_2D_steppables import InfoPrinterSteppable
infoPrinterSteppable=InfoPrinterSteppable(_simulator=sim,_frequency=10)
steppableRegistry.registerSteppable(infoPrinterSteppable)
CompuCellSetup.mainLoop(sim,simthread,steppableRegistry)
Now, all you need to do is to open in the Player previous xml file (cellsort_2D.xml)
together with newly created CellInfoPrinter.py and check Run Python check box. Notice
that you are not loading directly MyStappables.py file. The module you stored to this file
will be called from CellInfoPrinter.py.
Try running the simulation and see if you got any performance degradation. Probably
not, but by using Python you have saved yourself a lot of tedious C++ coding, not to
mention that you do not need to care about dependencies, compilation, etc..
Writing your next Python steppable will require much less effort as well, as you will
quickly discover that you will be using same basic code template all over again. Instead
of thinking how the code you are writing fits in the overall framework you will just
concentrate on it’s core functionality and leave the rest to CompuCell3D.
In case you wonder how this is all possible , it is due to Object Oriented programming.
Hopefully this short tutorial will encourage you to learn more of object oriented
programming. It is really worth the effort.
More Complicated Simulations – Adding Extra Attribute To a Cell
In CompuCell3D simulations each cell by default will have several attributes such as
volume, surface, centroids , target volumet, cell id etc.
One can write a plugin that attaches additional attributes to a cell during run time. Notice
that this does not require any recompilation.
It is by far the easiest to attach additional cell attribute in Python. Not only there is no
need to recompile anything, but the actual task takes one line of code:
pyAttributeAdder,listAdder=CompuCellSetup.attachListToCells(sim)
Above we told CompuCell3D to attach a Python list to each cell that will be produced by
the CompuCell3D kernel.
We can access this list very easily from Python level. Python list is dynamic data
structure that can grow or shrink and can hold arbitrary Python objects. Therefore by
attaching a list to each cell we effectively came up with a way to attach any cell attribute.
And everything takes place during run time…
Full listing of simulation where each cell gets extra attribute – a list:
import sys
from os import environ
from os import getcwd
import string
sys.path.append(environ["PYTHON_MODULE_PATH"])
sys.path.append(getcwd()+"/examples_PythonTutorial")
import CompuCellSetup
sim,simthread = CompuCellSetup.getCoreSimulationObjects()
#Create extra player fields here or add attributes
pyAttributeAdder,listAdder=CompuCellSetup.attachListToCells(sim)
CompuCellSetup.initializeSimulationObjects(sim,simthread)
#Add Python steppables here
steppableRegistry=CompuCellSetup.getSteppableRegistry()
#here we will add ExtraAttributeCellsort steppable
from cellsort_2D_steppables import ExtraAttributeCellsort
extraAttributeCellsort=ExtraAttributeCellsort(_simulator=sim,_frequency=10)
steppableRegistry.registerSteppable(extraAttributeCellsort)
from cellsort_2D_steppables import TypeSwitcherSteppable
typeSwitcherSteppable=TypeSwitcherSteppable(sim,100)
steppableRegistry.registerSteppable(typeSwitcherSteppable)
CompuCellSetup.mainLoop(sim,simthread,steppableRegistry)
ExtraAttributeCellsort
class ExtraAttributeCellsort(SteppablePy):
def __init__(self,_simulator,_frequency=10):
SteppablePy.__init__(self,_frequency)
self.simulator=_simulator
self.inventory=self.simulator.getPotts().getCellInventory()
self.cellList=CellList(self.inventory)
def step(self,mcs):
for cell in self.cellList:
pyAttrib=CompuCell.getPyAttrib(cell)
pyAttrib[0:2]=[cell.id*mcs , cell.id*(mcs-1)]
print "CELL ID modified=",pyAttrib[0]," ", pyAttrib[1]
Initializing first two elements of the list
Notice, you may also attach a dictionary to a cell instead of a list. See Python
Scripting Tutorials for more information. Dictionaries are actually more useful then lists in
the CompuCell3D context so make sure you understand them and know how to attach
them to cells.
TypeSwitcherSteppable
class TypeSwitcherSteppable(SteppablePy):
def __init__(self,_simulator,_frequency=100):
SteppablePy.__init__(self,_frequency)
self.simulator=_simulator
self.inventory=self.simulator.getPotts().getCellInventory()
self.cellList=CellList(self.inventory)
def step(self,mcs):
for cell in self.cellList:
if cell.type==1:
cell.type=2
elif (cell.type==2):
cell.type=1
else:
print "Unknown type. In cellsort simulation there should only be two types ”,
“1 and 2"
Line continuation in Python
Accessing NeighborTracker from Python
As you remember from lectures on XML CompuCell3D configuration files, CompuCell3D
can track cell neighbors. You can access information about cell neighbors directly from
Python:
class NeighborTrackerPrinterSteppable(SteppablePy):
def __init__(self,_simulator,_frequency=100):
SteppablePy.__init__(self,_frequency)
self.simulator=_simulator
self.nTrackerPlugin=CompuCell.getNeighborTrackerPlugin()
self.inventory=self.simulator.getPotts().getCellInventory()
self.cellList=CellList(self.inventory)
def start(self):pass
def step(self,mcs):
nTrackerAccessor=self.nTrackerPlugin.getNeighborTrackerAccessorPtr()
self.cellList=CellList(self.inventory)
for cell in self.cellList:
cellNeighborList=CellNeighborList(nTrackerAccessor,cell)
print "*********NEIGHBORS OF CELL WITH ID ",cell.id," *****************"
for neighborCell in cellNeighborList:
if neighborCell:
print "neighbor.id",neighborCell.id
Understanding iteration over cell neighbors
def step(self,mcs):
nTrackerAccessor=self.nTrackerPlugin.getNeighborTrackerAccessorPtr()
self.cellList=CellList(self.inventory)
for cell in self.cellList:
cellNeighborList=CellNeighborList(nTrackerAccessor,cell)
print "*********NEIGHBORS OF CELL WITH ID ",cell.id," *****************"
for neighborCell in cellNeighborList:
if neighborCell:
print "neighbor.id",neighborCell.id
This object “knows” how to
access cell neighbors
This function constructs list of
cell neighbors
Iterating over all cells in the simulation
Iterating over cell neighbors
import sys
from os import environ
from os import getcwd
import string
sys.path.append(environ["PYTHON_MODULE_PATH"])
sys.path.append(getcwd()+"/examples_PythonTutorial")
import CompuCellSetup
sim,simthread = CompuCellSetup.getCoreSimulationObjects()
#Create extra player fields here or add attributes
pyAttributeAdder,listAdder=CompuCellSetup.attachListToCells(sim)
CompuCellSetup.initializeSimulationObjects(sim,simthread)
#Add Python steppables here
steppableRegistry=CompuCellSetup.getSteppableRegistry()
from cellsort_2D_steppables import NeighborTrackerPrinterSteppable
neighborTrackerPrinterSteppable=NeighborTrackerPrinterSteppable(sim,100)
steppableRegistry.registerSteppable(neighborTrackerPrinterSteppable)
CompuCellSetup.mainLoop(sim,simthread,steppableRegistry)
Printing values of the concentration to a file
import sys
from os import environ
from os import getcwd
import string
sys.path.append(environ["PYTHON_MODULE_PATH"])
sys.path.append(getcwd()+"/examples_PythonTutorial")
import CompuCellSetup
sim,simthread = CompuCellSetup.getCoreSimulationObjects()
#Create extra player fields here or add attributes
CompuCellSetup.initializeSimulationObjects(sim,simthread)
#Add Python steppables here
from PySteppablesExamples import SteppableRegistry
steppableRegistry=SteppableRegistry()
from cellsort_2D_steppables import ConcentrationFieldDumperSteppable
concentrationFieldDumperSteppable=ConcentrationFieldDumperSteppable(sim,_frequency=100)
concentrationFieldDumperSteppable.setFieldName("FGF")
steppableRegistry.registerSteppable(concentrationFieldDumperSteppable)
CompuCellSetup.mainLoop(sim,simthread,steppableRegistry)
class ConcentrationFieldDumperSteppable(SteppablePy):
def __init__(self,_simulator,_frequency=1):
SteppablePy.__init__(self,_frequency)
self.simulator=_simulator
self.dim=self.simulator.getPotts().getCellFieldG().getDim()
def setFieldName(self,_fieldName):
self.fieldName=_fieldName
def step(self,mcs):
fileName=self.fieldName+"_"+str(mcs)+".dat"
self.outputField(self.fieldName,fileName)
def outputField(self,_fieldName,_fileName):
field=CompuCell.getConcentrationField(self.simulator,_fieldName)
pt=CompuCell.Point3D()
if field:
try:
fileHandle=open(_fileName,"w")
except IOError:
print "Could not open file ", _fileName," for writing. Check if you have necessary permissions"
for i in xrange(self.dim.x): #iterating over each lattice pixel
for j in xrange(self.dim.y):
for k in xrange(self.dim.z):
pt.x=i
pt.y=j
pt.z=k
fileHandle.write("%d\t%d\t%d\t%f\n"%(pt.x,pt.y,pt.z,field.get(pt)))
Creating, initializing and manipulating a concentration field directly from Python
•Although in most cases concentration fields are created and manipulated by PDE
solvers it is possible to accomplish all those tasks directly from Python.
•This can be very useful if you want to develop custom visualization that is not directly
supported by the Player. For example you may want to color cells according to how
many neighbors they have. Player does not offer such an option but you can implement
it very easily in Python in less then 5 minutes. This is not a joke. I am sure that by
combining two examples from this tutorial you will accomplish this task very fast.
The task of adding extra field to the Player and “managing” it consist of two steps
•Creating extra field and registering it with the Player and CompuCell3D kernel
•Writing steppable that manipulates values stored in the field
First let’s look at the full listing:
from os import environ
from os import getcwd
import string
sys.path.append(environ["PYTHON_MODULE_PATH"])
sys.path.append(getcwd()+"/examples_PythonTutorial")
import CompuCellSetup
sim,simthread = CompuCellSetup.getCoreSimulationObjects()
import CompuCell
Creating extra field is is really
easy. The location of the
function call that creates the
field is , however important. See
the comment
#Create extra player fields here or add attributes
dim=sim.getPotts().getCellFieldG().getDim()
extraPlayerField=simthread.createFloatFieldPy(dim,"ExtraField")
CompuCellSetup.initializeSimulationObjects(sim,simthread)
#Add Python steppables here
from PySteppablesExamples import SteppableRegistry
steppableRegistry=SteppableRegistry()
from cellsort_2D_steppables import ExtraFieldVisualizationSteppable
extraFieldVisualizationSteppable=ExtraFieldVisualizationSteppable(_simulator=sim,_frequency=10)
extraFieldVisualizationSteppable.setScalarField(extraPlayerField)
steppableRegistry.registerSteppable(extraFieldVisualizationSteppable)
CompuCellSetup.mainLoop(sim,simthread,steppableRegistry)
from PlayerPython import * # necessary to manipulate Player fields
from math import * # getting access to special functions from math module
class ExtraFieldVisualizationSteppable(SteppablePy):
def __init__(self,_simulator,_frequency=10):
SteppablePy.__init__(self,_frequency)
self.simulator=_simulator
self.cellFieldG=self.simulator.getPotts().getCellFieldG()
self.dim=self.cellFieldG.getDim()
def setScalarField(self,_field): # getting access to newly created field
self.scalarField=_field
def start(self):pass
def step(self,mcs):
for x in xrange(self.dim.x): #iteration over each pixel
for y in xrange(self.dim.y):
for z in xrange(self.dim.z):
pt=CompuCell.Point3D(x,y,z)
if (not mcs%20): #filling the values of the concentration
value=x*y # sometimes it is x*y
fillScalarValue(self.scalarField,x,y,z,value)
else:
value=sin(x*y) # sometimes sin(x*y)
fillScalarValue(self.scalarField,x,y,z,value)
Mitosis in CompuCell3D simulations
Supporting cell division (mitosis) in CompuCell3D simulations is a prerequisite for
building faithful biomedical simulations.
You can use mitosis module (Mitosis Plugin) directly from XML however, its use will be
very limited because of the following fact:
After cell division you end up with two cells. What parameters should those two cells
have (type, target volume etc.)? How do you modify the parameters?
The best solution is to manage mitosis from Python and the example below will explain
you how to do it.
import sys
from os import environ
from os import getcwd
import string
sys.path.append(environ["PYTHON_MODULE_PATH"])
sys.path.append(getcwd()+"/examples_PythonTutorial")
import CompuCellSetup
sim,simthread = CompuCellSetup.getCoreSimulationObjects()
import CompuCell #notice importing CompuCell to main script has to be done after call to
getCoreSimulationObjects()
from cellsort_2D_plugins import MitosisPyPlugin
changeWatcherRegistry=CompuCellSetup.getChangeWatcherRegistry(sim)
stepperRegistry=CompuCellSetup.getStepperRegistry(sim)
mitPy=MitosisPyPlugin(changeWatcherRegistry)
changeWatcherRegistry.registerPyChangeWatcher(mitPy) # registering mitosis as changeWatcher
stepperRegistry.registerPyStepper(mitPy) # registering mitosis as stepper
CompuCellSetup.initializeSimulationObjects(sim,simthread)
#Add Python steppables here
from PySteppablesExamples import SteppableRegistry
steppableRegistry=SteppableRegistry()
CompuCellSetup.mainLoop(sim,simthread,steppableRegistry)
Mitosis function is a type of plugin that monitors lattice (field3DWatcher) and a stepper
thus it need to implement two functions:
field3DChange
step
from CompuCell import MitosisSimplePlugin
class MitosisPyPlugin (StepperPy,Field3DChangeWatcherPy):
def __init__(self,_changeWatcher):
Field3DChangeWatcherPy.__init__(self,_changeWatcher)
#getting a handle to C++ mitosis module
self.mitosisPlugin=MitosisSimplePlugin()
self.doublingVolume=50 # setting critical volume that triggers mitosis to run
self.mitosisPlugin.setDoublingVolume(self.doublingVolume)
self.mitosisPlugin.turnOn()
# properly initializing C++ module
self.mitosisPlugin.init(self.changeWatcher.sim) self.counter=0
self.mitosisFlag=0
def setPotts(self,potts): #getting reference to Potts C++ object
self.mitosisPlugin.setPotts(potts)
Mitosis cont…
def field3DChange(self):# watching the lattice called only after spin copy
if self.changeWatcher.newCell and \
self.changeWatcher.newCell.volume>self.doublingVolume:
print "BIG CELL I WILL DO MITOSIS"
self.mitosisPlugin.field3DChange(self.changeWatcher.changePoint, \
self.changeWatcher.newCell, \
self.changeWatcher.newCell)
self.mitosisFlag=1
def step(self): # running actual mitosis
if self.mitosisFlag:
self.mitosisFlag=self.mitosisPlugin.doMitosis()
self.childCell=self.mitosisPlugin.getChildCell()
self.parentCell=self.mitosisPlugin.getParentCell()
self.updateAttributes()
self.mitosisFlag=0
def updateAttributes(self): # setting attributes of parent and daughter cells
self.childCell.targetVolume=self.parentCell.targetVolume
if self.parentCell.type==1:
self.childCell.type=2
else:
self.childCell.type=1
Mitosis was our first example of a plugin implemented in Python. We can implement
other plugins for example energy function in Python as well:
class VolumeEnergyFunctionPlugin(EnergyFunctionPy):
def __init__(self,_energyWrapper):# proper initialization
EnergyFunctionPy.__init__(self)
self.energyWrapper=_energyWrapper
self.vt=0.0
self.lambda_v=0.0
def setParams(self,_lambda,_targetVolume):# configuration of the plugin
self.lambda_v=_lambda;
self.vt=_targetVolume
def changeEnergy(self): # core function of energy function plugin
energy=0.0
if(self.energyWrapper.newCell):
energy+=self.lambda_v*(1+2*(self.energyWrapper.newCell.volume-self.vt))
if(self.energyWrapper.oldCell):
energy+=self.lambda_v*(1-2*(self.energyWrapper.oldCell.volume-self.vt))
return energy
Full script:
import sys
from os import environ
from os import getcwd
import string
sys.path.append(environ["PYTHON_MODULE_PATH"])
sys.path.append(getcwd()+"/examples_PythonTutorial")
import CompuCellSetup
sim,simthread = CompuCellSetup.getCoreSimulationObjects()
#Create extra player fields here or add attributes or plugins
energyFunctionRegistry=CompuCellSetup.getEnergyFunctionRegistry(sim)
from cellsort_2D_plugins import VolumeEnergyFunctionPlugin
volumeEnergy=VolumeEnergyFunctionPlugin(energyFunctionRegistry)
volumeEnergy.setParams(2.0,25.0)
energyFunctionRegistry.registerPyEnergyFunction(volumeEnergy)
CompuCellSetup.initializeSimulationObjects(sim,simthread)
#Add Python steppables here
steppableRegistry=CompuCellSetup.getSteppableRegistry()
CompuCellSetup.mainLoop(sim,simthread,steppableRegistry)
XML file
<CompuCell3D>
<Potts>
<Dimensions x="100" y="100" z="1"/>
<Steps>10000</Steps>
<Temperature>10</Temperature>
</Potts>
<Plugin Name="VolumeTracker"/> Notice we eliminated Volume plugin but need to keep VolumeTracker
Plugin
<Plugin Name="CellType">
<CellType TypeName="Medium" TypeId="0"/>
<CellType TypeName="Condensing" TypeId="1"/>
<CellType TypeName="NonCondensing" TypeId="2"/>
</Plugin>
<Plugin Name="Contact">
<Energy Type1="Medium" Type2="Medium">0</Energy>
<Energy Type1="NonCondensing" Type2="NonCondensing">16</Energy>
…
</Plugin>
<Steppable Type="BlobInitializer">
<Gap>0</Gap>
<Width>5</Width>
…
</Steppable>
</CompuCell3D>
Why CompuCell3D and why now?
Typical Evolution of Research Software
(in collaboration with Chris Mueller)
• First application written in Fortran
F77
• Tom ports to C
• Adds command line parameters, makefile
• Jenny ports to F90
• Extends model
C
F90
• Brad ports to C++
• Models system using objects
C++
C++
Java
• Maria implements simulation in Java for the Grid
• Implements original model
• Sounds familiar? Keep listening…
A Closer Look
F771
F772
C1 C3
C2
C++1
F901
F903 F902
F904
C++
Java
C++2
Version used for paper
Versions that advanced science
• There were 6 major versions
– 13 actual implementations
– 5 Languages
• 2 major versions advanced the science
• 4 major versions were simply software
projects
• All versions re-implemented basic features
• The implementations used for the papers
were not always used for the next major
version
Problem: Research software applications are difficult to develop and are costing
researchers time and money.
Solution: Separate Research and Development and use a development
model derived from industrial software development.
Benefits
• Software Quality is improved
– Applications are not single-user prototypes
– Features are available to all researchers and are peer-reviewed
• Research Process is improved
– Researchers can focus on research
– Development is not a bottleneck
– Reproducibility and Traceability
• Reproduce old experiments, trace the data/process that led to
a result
– Easier to integrate new/visiting researchers
• High-end software becomes possible
– Parallel and high-performance implementations
– Well designed user interfaces, visualization, databases, web
applications and services
CompuCell3D collaboration
University of Notre Dame , Indiana University
People:
Mark Alber, Ariel Balter, Rajiv Chaturvedi,Nan Chen, Trevor Cickovski, Jeff Coffland,
Michael Crocker, Gabor Forgacs, James Glazier, Tilmann Glimm, George
Hentschel, Chengbang Huang, Jesus Izaguirre, Chris Mueller, Stuart Newman,
Nikodem Poplawski, Maciej Swat, Gilberto Thomas
Funding agencies and institutions:
NSF, ICSB Notre Dame, The Biocomplexity Institute Indiana University, NIH
Users – most important part of the collaboration.
Website:
http://sourceforge.net/projects/compucell/
Google keyword: CompuCell3D
Running the simulation
Steering bar allows users to start or pause the
simulation, zoom in , zoom out, to switch between
2D and 3D visualization, change view modes (cell
field, pressure field , chemical concentration field,
velocity field etc..)
Player can output multiple views
during single simulation run –
Add Screenshot function
Information bar
Scripting languages – extending capabilities of CompuCell3D
BIOLOGO - high level, XML based language for describing parts of the simulation.
BIOLOGO is translated to C++ and compiled as a CompuCell3D plugin to deliver
optimal performance. It is straightforward to use
Kernel
Plugin
C++
BIOLOGO
Example:
Creating a PDE solver is reduced to writing several statements in BIOLOGO.
<DiffEq fieldname="C">
<Term exp="Kronecker(x,y,z)*alpha - epsilon*C*(1-Kronecker(x,y,z)) +
DiffConst*Laplacian(C)"
condition="true" />
</DiffEq>
Python:
We will embed Python in CompuCell3D to give users similar level of flexibility as Matlab
or Mathematica. Most of the low level functions will be implemented in C++ but callable
from Python level.
Example simulations using CompuCell3D
Dictyostelium discoideum
The model is based on oscilatory Fitzhugh-Nagumo type of equations
(Savill Hogeweg, 1997):
c
 D 2 c  f (c)  r
t
r
 (c)(kc  b  r )
t
c – denotes cAMP concentration - activator
r-refractoriness - inhibitor
f(c) – piecewise linear function
b=0 for excitable cells , b != 0 for autocycling cells
This system of equations will produce waves of cAMP passing
through the lattice. Cells respond to cAMP gradient by biasing
their movement towards higher level of cAMP. We assume that
cells chemotact “periodically” and only when cAMP
concentration is high enough
Dictyostelium Discoideum – early stages of aggregation
Autocycling amoebae
initiate waves of cAMP
cAMP wave passes
through amoebae and
a substrate
Amoebae move up the
cAMP concentration
gradient - chemotaxis
Mound is formed with cAMP wave propagates
autocycling cells at the down the mound
top
Cell sorting
Cell sorting is a rearrangement of cells driven by local physical properties of cells (cell-cell adhesion). Different parameters
lead to different final patterns:
Checkerboard pattern arises when
adhesion between red and green
cells is stronger than adhesion
between cells of the same type
Complete sorting – adhesion
between red cells is stronger then
between green cells and between
red and green
Initial configuration of condensing (red)
and non-condensing (green) cells
immersed in medium (blue)
Partial sorting – very similar to
complete sorting except that the
adhesion between red and green
cells is higher than adhesion
between green cells
Gastrulation
Ariel Balter
Area Opaca
Area Pellucida
Chemo-repellant source
Chemo-attractant source
Characteristic bulging due to
pressure from aggregating cells
All higher animals undergo a process called gastrulation at the earliest stages of development. During gastrulation
the embryo takes the first steps from a more-or-less uniform ball of cells towards a differentiated organism with
individual organs. One of the key process that begins gastrulation is the onset of large-scale motion of cells which
first deform the symmetric blastoderm (the stage preceding gastrulation). It has been hypothesized that cell motion
in response to chemical attractors and repellors plays an important role in this process.
Here we test a highly simplistic model of chemo-attraction/repulsion in the chick embryo. This model does capture
certain features of overall cell redistribution as well as the resulting remodeling of the embryo as a whole. However
it does not reproduce vortex motion observed in work done under James Glazier.
Fibronectin driven condensation as a function of cell density
Nikodem Poplawski
We are able to study how fibronectin-cell interaction and initial cell density affects cell
condensation patterns. Fibronectin limits the size of forming clusters.
High cell density results in
stripes of cell condensate
Medium cell density
produces a mixture of
condensate stripes and
spots
Low cell density leads to
condensate spots
Roadmap
CompuCell3D needs further development in order to become a universal
framework for modeling morphogenesis:
1st quarter 2006
-Full support for scripting languages
-Isolate users from XML - graphical configuration tools
-Provide full diagnostics of the simulation in CompuCell Player – for example rightclick should provide information about given cell
2nd-3rd quarter 2006
-Movies, scientific plots and all lattice information will be available through the
Player
-Parallel version
2006/2007
-Steering and parameters sweeps
-Serializing simulation – users will be able to stop the simulaiton and resume it later
-Project management