## Category: agent-based model

## Working with PhysiCell MultiCellDS digital snapshots in Matlab

PhysiCell 1.2.1 and later saves data as a specialized MultiCellDS digital snapshot, which includes chemical substrate fields, mesh information, and a readout of the cells and their phenotypes at single simulation time point. This tutorial will help you learn to use the matlab processing files included with PhysiCell.

This tutorial assumes you know (1) how to work at the shell / command line of your operating system, and (2) basic plotting and other functions in Matlab.

### Key elements of a PhysiCell digital snapshot

A PhysiCell digital snapshot (a customized form of the MultiCellDS digital simulation snapshot) includes the following elements saved as XML and MAT files:

**output12345678.xml**: This is the “base” output file, in MultiCellDS format. It includes key metadata such as when the file was created, the software, microenvironment information, and custom data saved at the simulation time. The Matlab files read this base file to find other related files (listed next).**Example**: output00003696.xml**initial_mesh0.mat**: This is the computational mesh information for BioFVM at time 0.0. Because BioFVM and PhysiCell do not use moving meshes, we do not save this data at any subsequent time.**output12345678_microenvironment0.mat**: This saves each biochemical substrate in the microenvironment at the computational voxels defined in the mesh (see above).**Example:**output00003696_microenvironment0.mat**output12345678_cells.mat**: This saves very basic cellular information related to BioFVM, including cell positions, volumes, secretion rates, uptake rates, and secretion saturation densities.**Example:**output00003696_cells.mat**output12345678_cells_physicell.mat**: This saves extra PhysiCell data for each cell agent, including volume information, cell cycle status, motility information, cell death information, basic mechanics, and any user-defined custom data.**Example:**output00003696_cells_physicell.mat

These snapshots make extensive use of Matlab Level 4 .mat files, for fast, compact, and well-supported saving of array data. Note that even if you cannot ready MultiCellDS XML files, you can work to parse the .mat files themselves.

### The PhysiCell Matlab .m files

Every PhysiCell distribution includes some matlab functions to work with PhysiCell digital simulation snapshots, stored in the matlab subdirectory. The main ones are:

**composite_cutaway_plot.m**: provides a quick, coarse 3-D cutaway plot of the discrete cells, with different colors for live (red), apoptotic (b), and necrotic (black) cells.**read_MultiCellDS_xml.m**: reads the “base” PhysiCell snapshot and its associated matlab files.**set_MCDS_constants.m**: creates a data structure MCDS_constants that has the same constants as PhysiCell_constants.h. This is useful for identifying cell cycle phases, etc.**simple_cutaway_plot.m**: provides a quick, coarse 3-D cutaway plot of user-specified cells.**simple_plot.m**: provides, a quick, coarse 3-D plot of the user-specified cells, without a cutaway or cross-sectional clipping plane.

#### A note on GNU Octave

Unfortunately, GNU octave does not include XML file parsing without some significant user tinkering. And one you’re done, it is approximately one order of magnitude slower than Matlab. Octave users can directly import the .mat files described above, but without the helpful metadata in the XML file. We’ll provide more information on the structure of these MAT files in a future blog post. Moreover, we plan to provide python and other tools for users without access to Matlab.

### A sample digital snapshot

We provide a 3-D simulation snapshot from the final simulation time of the cancer-immune example in Ghaffarizadeh et al. (2017, in review) at:

The corresponding SVG cross-section for that time (through *z *= 0 μm) looks like this:

Unzip the sample dataset in any directory, and make sure the matlab files above are in the same directory (or in your Matlab path). If you’re inside matlab:

!unzip 3D_PhysiCell_matlab_sample.zip

### Loading a PhysiCell MultiCellDS digital snapshot

Now, load the snapshot:

MCDS = read_MultiCellDS_xml( 'output00003696.xml');

This will load the mesh, substrates, and discrete cells into the MCDS data structure, and give a basic summary:

Typing ‘MCDS’ and then hitting ‘tab’ (for auto-completion) shows the overall structure of MCDS, stored as metadata, mesh, continuum variables, and discrete cells:

To get simulation metadata, such as the current simulation time, look at MCDS.metadata.current_time

Here, we see that the current simulation time is 30240 minutes, or 21 days. MCDS.metadata.current_runtime gives the elapsed walltime to up to this point: about 53 hours (1.9e5 seconds), including file I/O time to write full simulation data once per 3 simulated minutes after the start of the adaptive immune response.

### Plotting chemical substrates

Let’s make an oxygen contour plot through z = 0 μm. First, we find the index corresponding to this *z-*value:

k = find( MCDS.mesh.Z_coordinates == 0 );

Next, let’s figure out which variable is oxygen. Type “MCDS.continuum_variables.name”, which will show the array of variable names:

Here, oxygen is the first variable, (index 1). So, to make a filled contour plot:

contourf( MCDS.mesh.X(:,:,k), MCDS.mesh.Y(:,:,k), ... MCDS.continuum_variables(1).data(:,:,k) , 20 ) ;

Now, let’s set this to a correct aspect ratio (no stretching in x or y), add a colorbar, and set the axis labels, using

metadata to get labels:

axis image colorbar xlabel( sprintf( 'x (%s)' , MCDS.metadata.spatial_units) ); ylabel( sprintf( 'y (%s)' , MCDS.metadata.spatial_units) );

Lastly, let’s add an appropriate (time-based) title:

title( sprintf('%s (%s) at t = %3.2f %s, z = %3.2f %s', MCDS.continuum_variables(1).name , ... MCDS.continuum_variables(1).units , ... MCDS.metadata.current_time , ... MCDS.metadata.time_units, ... MCDS.mesh.Z_coordinates(k), ... MCDS.metadata.spatial_units ) );

Here’s the end result:

We can easily export graphics, such as to PNG format:

print( '-dpng' , 'output_o2.png' );

For more on plotting BioFVM data, see the tutorial

at http://www.mathcancer.org/blog/saving-multicellds-data-from-biofvm/

### Plotting cells in space

#### 3-D point cloud

First, let’s plot all the cells in 3D:

plot3( MCDS.discrete_cells.state.position(:,1) , MCDS.discrete_cells.state.position(:,2), ... MCDS.discrete_cells.state.position(:,3) , 'bo' );

At first glance, this does not look good: some cells are far out of the simulation domain, distorting the automatic range of the plot:

This does not ordinarily happen in PhysiCell (the default cell mechanics functions have checks to prevent such behavior), but this example includes a simple Hookean elastic adhesion model for immune cell attachment to tumor cells. In rare circumstances, an attached tumor cell or immune cell can apoptose on its own (due to its background apoptosis rate),

without “knowing” to detach itself from the surviving cell in the pair. The remaining cell attempts to calculate its elastic velocity based upon an invalid cell position (no longer in memory), creating an artificially large velocity that “flings” it out of the simulation domain. Such cells are not simulated any further, so this is effectively equivalent to an extra apoptosis event (only 3 cells are out of the simulation domain after tens of millions of cell-cell elastic adhesion calculations). Future versions of this example will include extra checks to prevent this rare behavior.

The plot can simply be fixed by changing the axis:

axis( 1000*[-1 1 -1 1 -1 1] ) axis square

Notice that this is a very difficult plot to read, and very non-interactive (laggy) to rotation and scaling operations. We can make a slightly nicer plot by searching for different cell types and plotting them with different colors:

% make it easier to work with the cell positions; P = MCDS.discrete_cells.state.position; % find type 1 cells ind1 = find( MCDS.discrete_cells.metadata.type == 1 ); % better still, eliminate those out of the simulation domain ind1 = find( MCDS.discrete_cells.metadata.type == 1 & ... abs(P(:,1))' < 1000 & abs(P(:,2))' < 1000 & abs(P(:,3))' < 1000 ); % find type 0 cells ind0 = find( MCDS.discrete_cells.metadata.type == 0 & ... abs(P(:,1))' < 1000 & abs(P(:,2))' < 1000 & abs(P(:,3))' < 1000 ); %now plot them P = MCDS.discrete_cells.state.position; plot3( P(ind0,1), P(ind0,2), P(ind0,3), 'bo' ) hold on plot3( P(ind1,1), P(ind1,2), P(ind1,3), 'ro' ) hold off axis( 1000*[-1 1 -1 1 -1 1] ) axis square

However, this isn’t much better. You can use the scatter3 function to gain more control on the size and color of the plotted cells, or even make macros to plot spheres in the cell locations (with shading and lighting), but Matlab is very slow when plotting beyond 10^{3} cells. Instead, we recommend the faster preview functions below for data exploration, and higher-quality plotting (e.g., by POV-ray) for final publication-

#### Fast 3-D cell data previewers

Notice that plot3 and scatter3 are painfully slow for any nontrivial number of cells. We can use a few fast previewers to quickly get a sense of the data. First, let’s plot all the dead cells, and make them red:

clf simple_plot( MCDS, MCDS, MCDS.discrete_cells.dead_cells , 'r' )

This function creates a coarse-grained 3-D indicator function (0 if no cells are present; 1 if they are), and plots a 3-D level surface. It is very responsive to rotations and other operations to explore the data. You may notice the second argument is a list of indices: only these cells are plotted. This gives you a method to select cells with specific characteristics when plotting. (More on that below.) If you want to get a sense of the interior structure, use a cutaway plot:

clf simple_cutaway_plot( MCDS, MCDS, MCDS.discrete_cells.dead_cells , 'r' )

We also provide a fast “composite” cutaway which plots all live cells as red, apoptotic cells as blue (without the cutaway), and all necrotic cells as black:

clf composite_cutaway_plot( MCDS )

Lastly, we show an improved plot that uses different colors for the immune cells, and Matlab’s “find” function to help set up the indexing:

constants = set_MCDS_constants % find the type 0 necrotic cells ind0_necrotic = find( MCDS.discrete_cells.metadata.type == 0 & ... (MCDS.discrete_cells.phenotype.cycle.current_phase == constants.necrotic_swelling | ... MCDS.discrete_cells.phenotype.cycle.current_phase == constants.necrotic_lysed | ... MCDS.discrete_cells.phenotype.cycle.current_phase == constants.necrotic) ); % find the live type 0 cells ind0_live = find( MCDS.discrete_cells.metadata.type == 0 & ... (MCDS.discrete_cells.phenotype.cycle.current_phase ~= constants.necrotic_swelling & ... MCDS.discrete_cells.phenotype.cycle.current_phase ~= constants.necrotic_lysed & ... MCDS.discrete_cells.phenotype.cycle.current_phase ~= constants.necrotic & ... MCDS.discrete_cells.phenotype.cycle.current_phase ~= constants.apoptotic) ); clf % plot live tumor cells red, in cutaway view simple_cutaway_plot( MCDS, ind0_live , 'r' ); hold on % plot dead tumor cells black, in cutaway view simple_cutaway_plot( MCDS, ind0_necrotic , 'k' ) % plot all immune cells, but without cutaway (to show how they infiltrate) simple_plot( MCDS, ind1, 'g' ) hold off

### A small cautionary note on future compatibility

PhysiCell 1.2.1 uses the <custom> data tag (allowed as part of the MultiCellDS specification) to encode its cell data, to allow a more compact data representation, because the current PhysiCell daft does not support such a formulation, and Matlab is * painfully *slow at parsing XML files larger than ~50 MB. Thus, PhysiCell snapshots are not yet fully compatible with general MultiCellDS tools, which would by default ignore custom data. In the future, we will make available converter utilities to transform “native” custom PhysiCell snapshots to MultiCellDS snapshots that encode all the cellular information in a more verbose but compatible XML format.

### Closing words and future work

Because Octave is not a great option for parsing XML files (with critical MultiCellDS metadata), we plan to write similar functions to read and plot PhysiCell snapshots in Python, as an open source alternative. Moreover, our lab in the next year will focus on creating further MultiCellDS configuration, analysis, and visualization routines. We also plan to provide additional 3-D functions for plotting the discrete cells and varying color with their properties.

In the longer term, we will develop open source, stand-alone analysis and visualization tools for MultiCellDS snapshots (including PhysiCell snapshots). Please stay tuned!

## Running the PhysiCell sample projects

### Introduction

In PhysiCell 1.2.1 and later, we include four sample projects on cancer heterogeneity, bioengineered multicellular systems, and cancer immunology. This post will walk you through the steps to build and run the examples.

If you are new to PhysiCell, you should first make sure you’re ready to run it. (Please note that this applies in particular for OSX users, as Xcode’s g++ is not compatible out-of-the-box.) Here are tutorials on getting ready to Run PhysiCell:

- Setting up a 64-bit gcc environment in Windows.
- Setting up gcc / OpenMP on OSX (MacPorts edition)
- Setting up gcc / OpenMP on OSX (Homebrew edition)

**Note:**This is the preferred method for Mac OSX. - Getting started with a PhysiCell Virtual Appliance (for virtual machines like VirtualBox)

**Note:**The “native” setups above are preferred, but the Virtual Appliance is a great “plan B” if you run into trouble

Please note that we expect to expand this tutorial.

### Building, running, and viewing the sample projects

All of these projects will create data of the following forms:

- Scalable vector graphics (SVG) cross-section plots through
*z*= 0.0 μm at each output time. Filenames will look like snapshot00000000.svg. - Matlab (Level 4) .mat files to store raw BioFVM data. Filenames will look like output00000000_microenvironment0.mat (for the chemical substrates) and output00000000_cells.mat (for basic agent data).
- Matlab .mat files to store additional PhysiCell agent data. Filenames will look like output00000000_cells_physicell.mat.
- MultiCellDS .xml files that give further metadata and structure for the .mat files. Filenames will look like output00000000.xml.

You can read the combined data in the XML and MAT files with the read_MultiCellDS_xml function, stored in the matlab directory of every PhysiCell download. (Copy the read_MultiCellDS_xml.m and set_MultiCelLDS_constants.m files to the same directory as your data for the greatest simplicity.)

(If you are using Mac OSX and PhysiCell version > 1.2.1, remember to set the PHYSICELL_CPP environment variable to be an OpenMP-capable compiler – rf. Homebrew setup.)

#### Biorobots (2D)

Type the following from a terminal window in your root PhysiCell directory:

make biorobots-sample make ./biorobots make reset # optional -- gets a clean slate to try other samples

Because this is a 2-D example, the SVG snapshot files will provide the simplest method of visualizing these outputs. You can use utilities like ImageMagick to convert them into other formats for publications, such as PNG or EPS.

#### Anti-cancer biorobots (2D)

make cancer-biorobots-sample make ./cancer_biorobots make reset # optional -- gets a clean slate to try other samples

#### Cancer heterogeneity (2D)

make heterogeneity-sample make project ./heterogeneity make reset # optional -- gets a clean slate to try other samples

#### Cancer immunology (3D)

make cancer-immune-sample make ./cancer_immune_3D make reset # optional -- gets a clean slate to try other samples

## Getting started with a PhysiCell Virtual Appliance

**Note:** This is part of a series of “how-to” blog posts to help new users and developers of BioFVM and PhysiCell. This guide is for for users in OSX, Linux, or Windows using the VirtualBox virtualization software to run a PhysiCell virtual appliance.

These instructions should get you up and running without needed to install a compiler, makefile capabilities, or any other software (beyond the virtual machine and the PhysiCell virtual appliance). We note that using the PhysiCell source with your own compiler is still the preferred / ideal way to get started, but the virtual appliance option is a fast way to start even if you’re having troubles setting up your development environment.

### What’s a Virtual Machine? What’s a Virtual Appliance?

A virtual machine is a full simulated computer (with its own disk space, operating system, etc.) running on another. They are designed to let a user test on a completely different environment, without affecting the host (main) environment. They also allow a very robust way of taking and reproducing the state of a full working environment.

A virtual appliance is just this: a full image of an installed system (and often its saved state) on a virtual machine, which can easily be installed on a new virtual machine. In this tutorial, you will download our PhysiCell virtual appliance and use its pre-configured compiler and other tools.

### What you’ll need:

**VirtualBox:**This is a free, cross-platform program to run virtual machines on OSX, Linux, Windows, and other platforms. It is a safe and easy way to install one full operating (a client system) on your main operating system (the host system). For us, this means that we can distribute a fully working Linux environment with a working copy of all the tools you need to compile and run PhysiCell. As of August 1, 2017, this will download Version 5.1.26.**Download here:**https://www.virtualbox.org/wiki/Downloads

**PhysiCell Virtual Appliance**: This is a single-file distribution of a virtual machine running Alpine Linux, including all key tools needed to compile and run PhysiCell. As of July 31, 2017, this will download PhysiCell 1.2.2 with g++ 6.3.0.**Download here:**https://sourceforge.net/projects/physicell/files/PhysiCell/- (Browse to a version of PhysiCell, and download the file that ends in “.ova”.)

**Version 1.2.0:**http://bit.ly/2vY51P1 [sf.net]

**A computer with hardware support for virtualization:**Your CPU needs to have hardware support for virtualization (almost all of them do now), and it has to be enabled in your BIOS. Consult your computer maker on how to turn this on if you get error messages later.

### Main steps:

#### 1) Install VirtualBox.

Double-click / open the VirtualBox download. Go ahead and accept all the default choices. If asked, go ahead and download/install the extensions pack.

#### 2) Import the PhysiCell Virtual Appliance

Go the “File” menu and choose “Import Virtual Appliance”. Browse to find the .ova file you just downloaded.

Click on “Next,” and import with all the default options. That’s it!

#### 3) [Optional] Change settings

You most likely won’t need this step, but you can increase/decrease the amount of RAM used for the virtual machine if you select the PhysiCell VM, click the Settings button (orange gear), and choose “System”:We set the Virtual Machine to have 4 GB of RAM. If you have a machine with lots of RAM (16 GB or more), you may want to set this to 8 GB.

Also, you can choose how many virtual CPUs to give to your VM:

We selected 4 when we set up the Virtual Appliance, but you should match the number of * physical processor cores *on your machine. In my case, I have a quad core processor with hyperthreading. This means 4 real cores, 8 virtual cores, so I select 4 here.

#### 4) Start the Virtual Machine and log in

Select the PhysiCell machine, and click the green “start” button. After the virtual machine boots (with the good old LILO boot manager that I’ve missed), you should see this:

Click the "More ..." button, and log in withusername:physicell,password:physicell

#### 5) Test the compiler and run your first simulation

Notice that PhysiCell is already there on the desktop in the PhysiCell folder. Right-click, and choose “open terminal here.” You’ll already be in the main PhysiCell root directory.

Now, let’s compile your first project! Type “make template2D && make” And run your project! Type “./project” and let it go!Go ahead and run either the first few days of the simulation (until about 7200 minutes), then hit <control>-C to cancel out. Or run the whole simulation–that’s fine, too.

#### 6) Look at the results

We bundled a few tools to easily look at results. First, * ristretto *is a very fast image viewer. Let’s view the SVG files: As a nice tip, you can press the left and right arrows to advance through the SVG images, or hold the right arrow down to advance through quickly.

Now, let’s use * ImageMagick* to convert the SVG files into JPG file: call “magick mogrify -format jpg snap*.svg”

Next, let’s turn those images into a movie. I generally create moves that are 24 frames pers se, so that 1 second of the movie is 1 hour of simulations time. We’ll use * mencoder*, with options below given to help get a good quality vs. size tradeoff:

When you’re done, view the movie with * mplayer*. The options below scale the window to fit within the virtual monitor:

If you want to loop the movie, add “-loop 999” to your command.

#### 7) Get familiar with other tools

Use * nano *(useage:

**nano <filename>**) to quickly change files at the command line. Hit <control>-O to save your results. Hit <control>-X to exit. <control>-W will search within the file.

Use * nedit *(useage:

**nedit <filename> &**) to open up one more text files in a graphical editor. This is a good way to edit multiple files at once.

Sometimes, you need to run commands at elevated (admin or root) privileges. Use ** sudo**. Here’s an example, searching the Alpine Linux package manager

*for clang:*

**apk**physicell:~$ sudo apk search gcc [sudo] password for physicell: physicell:~$ sudo apk search clang clang-analyzer-4.0.0-r0 clang-libs-4.0.0-r0 clang-dev-4.0.0-r0 clang-static-4.0.0-r0 emscripten-fastcomp-1.37.10-r0 clang-doc-4.0.0-r0 clang-4.0.0-r0 physicell:~/Desktop/PhysiCell$

If you want to install clang/llvm (as an alternative compiler):

physicell:~$ sudo apk add gcc [sudo] password for physicell: physicell:~$ sudo apk search clang clang-analyzer-4.0.0-r0 clang-libs-4.0.0-r0 clang-dev-4.0.0-r0 clang-static-4.0.0-r0 emscripten-fastcomp-1.37.10-r0 clang-doc-4.0.0-r0 clang-4.0.0-r0 physicell:~/Desktop/PhysiCell$

Notice that it asks for a password: use the password for root (which is physicell).

#### 8) [Optional] Configure a shared folder

Coming soon.

### Why both with zipped source, then?

Given that we can get a whole development environment by just downloading and importing a virtual appliance, why

bother with all the setup of a native development environment, like this tutorial (Windows) or this tutorial (Mac)?

One word: **performance**. In my testing, I still have not found the performance running inside a

virtual machine to match compiling and running directly on your system. So, the Virtual Appliance is a great

option to get up and running quickly while trying things out, but I still recommend setting up natively with

one of the tutorials I linked in the preceding paragraphs.

### What’s next?

In the coming weeks, we’ll post further tutorials on using PhysiCell. In the meantime, have a look at the

PhysiCell project website, and these links as well:

- BioFVM on MathCancer.org: http://BioFVM.MathCancer.org
- BioFVM on SourceForge: http://BioFVM.sf.net
- BioFVM Method Paper in BioInformatics: http://dx.doi.org/10.1093/bioinformatics/btv730
- PhysiCell on MathCancer.org: http://PhysiCell.MathCancer.org
- PhysiCell on Sourceforge: http://PhysiCell.sf.net
- PhysiCell Method Paper (preprint): https://doi.org/10.1101/088773
- PhysiCell tutorials: [click here]

## Coarse-graining discrete cell cycle models

### Introduction

One observation that often goes underappreciated in computational biology discussions is that a computational model is often * a model of a model of a model of biology*: that is, it’s a numerical approximation (a model) of a mathematical model of an experimental model of a real-life biological system. Thus, there are three big places where a computational investigation can fall flat:

- The experimental model may be a bad choice for the disease or process (not our fault).
- Second, the mathematical model of the experimental system may have flawed assumptions (something we have to evaluate).
- The numerical implementation may have bugs or otherwise be mathematically inconsistent with the mathematical model.

Critically, you can’t use simulations to evaluate the experimental model or the mathematical model until you verify that the numerical implementation is consistent with the mathematical model, and that the numerical solution converges as \( \Delta t\) and \( \Delta x \) shrink to zero.

There are numerous ways to accomplish this, but ideally, it boils down to having some analytical solutions to the mathematical model, and comparing numerical solutions to these analytical or theoretical results. In this post, we’re going to walk through the math of analyzing a typical type of discrete cell cycle model.

### Discrete model

Suppose we have a cell cycle model consisting of phases \(P_1, P_2, \ldots P_n \), where cells in the \(P_i\) phase progress to the \(P_{i+1}\) phase after a mean waiting time of \(T_i\), and cells leaving the \(P_n\) phase divide into two cells in the \(P_1\) phase. Assign each cell agent \(k\) a current phenotypic phase \( S_k(t) \). Suppose also that each phase \( i \) has a death rate \( d_i \), and that cells persist for on average \( T_\mathrm{A} \) time in the dead state before they are removed from the simulation.

The mean waiting times \( T_i \) are equivalent to transition rates \( r_i = 1 / T_i \) (Macklin et al. 2012). Moreover, for any time interval \( [t,t+\Delta t] \), both are equivalent to a transition probability of

\[ \mathrm{Prob}\Bigl( S_k(t+\Delta t) = P_{i+1} | S(t) = P_i \Bigr) = 1 – e^{ -r_i \Delta t } \approx r_i \Delta t = \frac{ \Delta t}{ T_i}. \] In many discrete models (especially cellular automaton models) with fixed step sizes \( \Delta t \), models are stated in terms of transition probabilities \( p_{i,i+1} \), which we see are equivalent to the work above with \( p_{i,i+1} = r_i \Delta t = \Delta t / T_i \), allowing us to tie mathematical model forms to biological, measurable parameters. We note that each \(T_i\) is the average duration of the \( P_i \) phase.

#### Concrete example: a Ki67 Model

Ki-67 is a nuclear protein that is expressed through much of the cell cycle, including S, G_{2}, M, and part of G_{1} after division. It is used very commonly in pathology to assess proliferation, particularly in cancer. See the references and discussion in (Macklin et al. 2012). In Macklin et al. (2012), we came up with a discrete cell cycle model to match Ki-67 data (along with cleaved Caspase-3 stains for apoptotic cells). Let’s summarize the key parts here.

Each cell agent \(i\) has a phase \(S_i(t)\). Ki67- cells are quiescent (phase \(Q\), mean duration \( T_\mathrm{Q} \)), and they can enter the Ki67+ \(K_1\) phase (mean duration \(T_1\)). When \( K_1 \) cells leave their phase, they divide into two Ki67+ daughter cells in the \( K_2 \) phase with mean duration \( T_2 \). When cells exit \( K_2 \), they return to \( Q \). Cells in any phase can become apoptotic (enter the \( A \) phase with mean duration \( T_\mathrm{A} \)), with death rate \( r_\mathrm{A} \).

### Coarse-graining to an ODE model

If each phase \(i\) has a death rate \(d_i\), if \( N_i(t) \) denotes the number of cells in the \( P_i \) phase at time \( t\), and if \( A(t) \) is the number of dead (apoptotic) cells at time \( t\), then on average, the number of cells in the \( P_i \) phase at the next time step is given by

\[ N_i(t+\Delta t) = N_i(t) + N_{i-1}(t) \cdot \left[ \textrm{prob. of } P_{i-1} \rightarrow P_i \textrm{ transition} \right] – N_i(t) \cdot \left[ \textrm{prob. of } P_{i} \rightarrow P_{i+1} \textrm{ transition} \right] \] \[ – N_i(t) \cdot \left[ \textrm{probability of death} \right] \] By the work above, this is:

\[ N_i(t+\Delta t) \approx N_i(t) + N_{i-1}(t) r_{i-1} \Delta t – N_i(t) r_i \Delta t – N_i(t) d_i \Delta t , \] or after shuffling terms and taking the limit as \( \Delta t \downarrow 0\), \[ \frac{d}{dt} N_i(t) = r_{i-1} N_{i-1}(t) – \left( r_i + d_i \right) N_i(t). \] Continuing this analysis, we obtain a linear system:

\[ \frac{d}{dt}{ \vec{N} } = \begin{bmatrix} -(r_1+d_1) & 0 & \cdots & 0 & 2r_n & 0 \\ r_1 & -(r_2+d_2) & 0 & \cdots & 0 & 0 \\ 0 & r_2 & -(r_3+d_3) & 0 & \cdots & 0 \\ & & \ddots & & \\0&\cdots&0 &r_{n-1} & -(r_n+d_n) & 0 \\ d_1 & d_2 & \cdots & d_{n-1} & d_n & -\frac{1}{T_\mathrm{A}} \end{bmatrix}\vec{N} = M \vec{N}, \] where \( \vec{N}(t) = [ N_1(t), N_2(t) , \ldots , N_n(t) , A(t) ] \).

For the Ki67 model above, let \(\vec{N} = [K_1, K_2, Q, A]\). Then the linear system is

\[ \frac{d}{dt} \vec{N} = \begin{bmatrix} -\left( \frac{1}{T_1} + r_\mathrm{A} \right) & 0 & \frac{1}{T_\mathrm{Q}} & 0 \\ \frac{2}{T_1} & -\left( \frac{1}{T_2} + r_\mathrm{A} \right) & 0 & 0 \\ 0 & \frac{1}{T_2} & -\left( \frac{1}{T_\mathrm{Q}} + r_\mathrm{A} \right) & 0 \\ r_\mathrm{A} & r_\mathrm{A} & r_\mathrm{A} & -\frac{1}{T_\mathrm{A}} \end{bmatrix} \vec{N} .\]

(If we had written \( \vec{N} = [Q, K_1, K_2 , A] \), then the matrix above would have matched the general form.)

### Some theoretical results

If \( M\) has eigenvalues \( \lambda_1 , \ldots \lambda_{n+1} \) and corresponding eigenvectors \( \vec{v}_1, \ldots , \vec{v}_{n+1} \), then the general solution is given by

\[ \vec{N}(t) = \sum_{i=1}^{n+1} c_i e^{ \lambda_i t } \vec{v}_i ,\] and if the initial cell counts are given by \( \vec{N}(0) \) and we write \( \vec{c} = [c_1, \ldots c_{n+1} ] \), we can obtain the coefficients by solving \[ \vec{N}(0) = [ \vec{v}_1 | \cdots | \vec{v}_{n+1} ]\vec{c} .\] In many cases, it turns out that all but one of the eigenvalues (say \( \lambda \) with corresponding eigenvector \(\vec{v}\)) are negative. In this case, all the other components of the solution decay away, and for long times, we have \[ \vec{N}(t) \approx c e^{ \lambda t } \vec{v} .\] This is incredibly useful, because it says that over long times, the fraction of cells in the \( i^\textrm{th} \) phase is given by \[ v_{i} / \sum_{j=1}^{n+1} v_{j}. \]

#### Matlab implementation (with the Ki67 model)

First, let’s set some parameters, to make this a little easier and reusable.

parameters.dt = 0.1; % 6 min = 0.1 hours parameters.time_units = 'hour'; parameters.t_max = 3*24; % 3 days parameters.K1.duration = 13; parameters.K1.death_rate = 1.05e-3; parameters.K1.initial = 0; parameters.K2.duration = 2.5; parameters.K2.death_rate = 1.05e-3; parameters.K2.initial = 0; parameters.Q.duration = 74.35 ; parameters.Q.death_rate = 1.05e-3; parameters.Q.initial = 1000; parameters.A.duration = 8.6; parameters.A.initial = 0;

Next, we write a function to read in the parameter values, construct the matrix (and all the data structures), find eigenvalues and eigenvectors, and create the theoretical solution. It also finds the positive eigenvalue to determine the long-time values.

function solution = Ki67_exact( parameters ) % allocate memory for the main outputs solution.T = 0:parameters.dt:parameters.t_max; solution.K1 = zeros( 1 , length(solution.T)); solution.K2 = zeros( 1 , length(solution.T)); solution.K = zeros( 1 , length(solution.T)); solution.Q = zeros( 1 , length(solution.T)); solution.A = zeros( 1 , length(solution.T)); solution.Live = zeros( 1 , length(solution.T)); solution.Total = zeros( 1 , length(solution.T)); % allocate memory for cell fractions solution.AI = zeros(1,length(solution.T)); solution.KI1 = zeros(1,length(solution.T)); solution.KI2 = zeros(1,length(solution.T)); solution.KI = zeros(1,length(solution.T)); % get the main parameters T1 = parameters.K1.duration; r1A = parameters.K1.death_rate; T2 = parameters.K2.duration; r2A = parameters.K2.death_rate; TQ = parameters.Q.duration; rQA = parameters.Q.death_rate; TA = parameters.A.duration; % write out the mathematical model: % d[Populations]/dt = Operator*[Populations] Operator = [ -(1/T1 +r1A) , 0 , 1/TQ , 0; ... 2/T1 , -(1/T2 + r2A) ,0 , 0; ... 0 , 1/T2 , -(1/TQ + rQA) , 0; ... r1A , r2A, rQA , -1/TA ]; % eigenvectors and eigenvalues [V,D] = eig(Operator); eigenvalues = diag(D); % save the eigenvectors and eigenvalues in case you want them. solution.V = V; solution.D = D; solution.eigenvalues = eigenvalues; % initial condition VecNow = [ parameters.K1.initial ; parameters.K2.initial ; ... parameters.Q.initial ; parameters.A.initial ] ; solution.K1(1) = VecNow(1); solution.K2(1) = VecNow(2); solution.Q(1) = VecNow(3); solution.A(1) = VecNow(4); solution.K(1) = solution.K1(1) + solution.K2(1); solution.Live(1) = sum( VecNow(1:3) ); solution.Total(1) = sum( VecNow(1:4) ); solution.AI(1) = solution.A(1) / solution.Total(1); solution.KI1(1) = solution.K1(1) / solution.Total(1); solution.KI2(1) = solution.K2(1) / solution.Total(1); solution.KI(1) = solution.KI1(1) + solution.KI2(1); % now, get the coefficients to write the analytic solution % [Populations] = c1*V(:,1)*exp( d(1,1)*t) + c2*V(:,2)*exp( d(2,2)*t ) + % c3*V(:,3)*exp( d(3,3)*t) + c4*V(:,4)*exp( d(4,4)*t ); coeff = linsolve( V , VecNow ); % find the (hopefully one) positive eigenvalue. % eigensolutions with negative eigenvalues decay, % leaving this as the long-time behavior. eigenvalues = diag(D); n = find( real( eigenvalues ) &gt; 0 ) solution.long_time.KI1 = V(1,n) / sum( V(:,n) ); solution.long_time.KI2 = V(2,n) / sum( V(:,n) ); solution.long_time.QI = V(3,n) / sum( V(:,n) ); solution.long_time.AI = V(4,n) / sum( V(:,n) ) ; solution.long_time.KI = solution.long_time.KI1 + solution.long_time.KI2; % now, write out the solution at all the times for i=2:length( solution.T ) % compact way to write the solution VecExact = real( V*( coeff .* exp( eigenvalues*solution.T(i) ) ) ); solution.K1(i) = VecExact(1); solution.K2(i) = VecExact(2); solution.Q(i) = VecExact(3); solution.A(i) = VecExact(4); solution.K(i) = solution.K1(i) + solution.K2(i); solution.Live(i) = sum( VecExact(1:3) ); solution.Total(i) = sum( VecExact(1:4) ); solution.AI(i) = solution.A(i) / solution.Total(i); solution.KI1(i) = solution.K1(i) / solution.Total(i); solution.KI2(i) = solution.K2(i) / solution.Total(i); solution.KI(i) = solution.KI1(i) + solution.KI2(i); end return;

Now, let’s run it and see what this thing looks like:

Next, we plot KI_{1}, KI_{2}, and AI versus time (solid curves), along with the theoretical long-time behavior (dashed curves). Notice how well it matches–it’s neat when theory works! 🙂

Some readers may recognize the long-time fractions: KI_{1} + KI_{2} = KI = 0.1743, and AI = 0.00833, very close to the DCIS patient values from our simulation study in Macklin et al. (2012) and improved calibration work in Hyun and Macklin (2013).

### Comparing simulations and theory

I wrote a small Matlab program to implement the discrete model: start with 1000 cells in the \(Q\) phase, and in each time interval \([t,t+\Delta t]\), each cell “decides” whether to advance to the next phase, stay in the same phase, or apoptose. If we compare a single run against the theoretical curves, we see hints of a match:

If we average 10 simulations and compare, the match is better:

And lastly, if we average 100 simulations and compare, the curves are very difficult to tell apart:

Even in logarithmic space, it’s tough to tell these apart:

### Code

The following matlab files (available here) can be used to reproduce this post:

- Ki67_exact.m
- The function defined above to create the exact solution using the eigenvalue/eignvector approach.
- Ki67_stochastic.m
- Runs a single stochastic simulation, using the supplied parameters.
- script.m
- Runs the theoretical solution first, creates plots, and then runs the stochastic model 100 times for comparison.

To make it all work, simply run “script” at the command prompt. Please note that it will generate some png files in its directory.

### Closing thoughts

In this post, we showed a nice way to check a discrete model against theoretical behavior–both in short-term dynamics and long-time behavior. The same work should apply to validating many discrete models. However, when you add spatial effects (e.g., a cellular automaton model that won’t proliferate without an empty neighbor site), I wouldn’t expect a match. (But simulating cells that initially have a “salt and pepper”, random distribution should match this for early times.)

Moreover, models with deterministic phase durations (e.g., K_{1}, K_{2}, and A have fixed durations) aren’t consistent with the ODE model above, unless the cells they are each initialized with a random amount of “progress” in their initial phases. (Otherwise, the cells in each phase will run synchronized, and there will be fixed delays before cells transition to other phases.) Delay differential equations better describe such models. However, for long simulation times, the slopes of the sub-populations and the cell fractions should start to better and better match the ODE models.

Now that we have verified that the discrete model is performing as expected, we can have greater confidence in its predictions, and start using those predictions to assess the underlying models. In ODE and PDE models, you often validate the code on simpler problems where you have an analytical solution, and then move on to making simulation predictions in cases where you can’t solve analytically. Similarly, we can now move on to variants of the discrete model where we can’t as easily match ODE theory (e.g., time-varying rate parameters, spatial effects), but with the confidence that the phase transitions are working as they should.

## Building a Cellular Automaton Model Using BioFVM

**Note:** This is part of a series of “how-to” blog posts to help new users and developers of BioFVM. See below for guides to setting up a C++ compiler in Windows or OSX.

### What you’ll need

- A working C++ development environment with support for OpenMP. See these prior tutorials if you need help.
- A download of BioFVM, available at http://BioFVM.MathCancer.org and http://BioFVM.sf.net. Use Version 1.1.4 or later.
- The source code for this project (see below).

Matlab or Octave for visualization. Matlab might be available for free at your university. Octave is open source and available from a variety of sources.

### Our modeling task

We will implement a basic 3-D cellular automaton model of tumor growth in a well-mixed fluid, containing oxygen pO_{2} (mmHg) and a drug *c* (e.g., doxorubicin, μM), inspired by modeling by Alexander Anderson, Heiko Enderling, Jan Poleszczuk, Gibin Powathil, and others. (I highly suggest seeking out the sophisticated cellular automaton models at Moffitt’s Integrated Mathematical Oncology program!) This example shows you how to extend BioFVM into a new cellular automaton model. I’ll write a similar post on how to add BioFVM into an existing cellular automaton model, which you may already have available.

Tumor growth will be driven by oxygen availability. Tumor cells can be live, apoptotic (going through energy-dependent cell death, or necrotic (undergoing death from energy collapse). Drug exposure can both trigger apoptosis and inhibit cell cycling. We will model this as growth into a well-mixed fluid, with pO_{2} = 38 mmHg (about 5% oxygen: a physioxic value) and *c* = 5 μM.

### Mathematical model

As a cellular automaton model, we will divide 3-D space into a regular lattice of voxels, with length, width, and height of 15 μm. (A typical breast cancer cell has radius around 9-10 μm, giving a typical volume around 3.6×10^{3} μm^{3}. If we make each lattice site have the volume of one cell, this gives an edge length around 15 μm.)

In voxels unoccupied by cells, we approximate a well-mixed fluid with Dirichlet nodes, setting pO_{2} = 38 mmHg, and initially setting *c *= 0. Whenever a cell dies, we replace it with an empty automaton, with no Dirichlet node. Oxygen and drug follow the typical diffusion-reaction equations:

\[ \frac{ \partial \textrm{pO}_2 }{\partial t} = D_\textrm{oxy} \nabla^2 \textrm{pO}_2 – \lambda_\textrm{oxy} \textrm{pO}_2 – \sum_{ \textrm{cells} i} U_{i,\textrm{oxy}} \textrm{pO}_2 \]

\[ \frac{ \partial c}{ \partial t } = D_c \nabla^2 c – \lambda_c c – \sum_{\textrm{cells }i} U_{i,c} c \]

where each uptake rate is applied across the cell’s volume. We start the treatment by setting *c = *5 μM on all Dirichlet nodes at *t *= 504 hours (21 days). For simplicity, we do not model drug degradation (pharmacokinetics), to approximate the *in vitro *conditions.

In any time interval [*t*,*t*+Δ*t*], each live tumor cell *i* has a probability *p _{i,D}* of attempting division, probability

*p*of apoptotic death, and probability

_{i,A}*p*of necrotic death. (For simplicity, we ignore motility in this version.) We relate these to the birth rate

_{i,N}*b*, apoptotic death rate

_{i}*d*, and necrotic death rate

_{i,A}*d*by the linearized equations (from Macklin et al. 2012):

_{i,N}\[ \textrm{Prob} \Bigl( \textrm{cell } i \textrm{ becomes apoptotic in } [t,t+\Delta t] \Bigr) = 1 – \textrm{exp}\Bigl( -d_{i,A}(t) \Delta t\Bigr) \approx d_{i,A}\Delta t \]

\[ \textrm{Prob} \Bigl( \textrm{cell } i \textrm{ attempts division in } [t,t+\Delta t] \Bigr) = 1 – \textrm{exp}\Bigl( -b_i(t) \Delta t\Bigr) \approx b_{i}\Delta t \]

\[ \textrm{Prob} \Bigl( \textrm{cell } i \textrm{ becomes necrotic in } [t,t+\Delta t] \Bigr) = 1 – \textrm{exp}\Bigl( -d_{i,N}(t) \Delta t\Bigr) \approx d_{i,N}\Delta t \]

*T*, which will vary by the type of cell death. Each dead cell automaton has a probability

_{i,D}*p*of lysis (rupture and removal) in any time span [

_{i,L}*t*,

*t+*Δ

*t*]. The duration

*T*is converted to a probability of cell lysis by

_{D}\[ \textrm{Prob} \Bigl( \textrm{dead cell } i \textrm{ lyses in } [t,t+\Delta t] \Bigr) = 1 – \textrm{exp}\Bigl( -\frac{1}{T_{i,D}} \Delta t\Bigr) \approx \frac{ \Delta t}{T_{i,D}} \]

##### (Illustrative) parameter values

We use *D*_{oxy} = 10^{5} μm^{2}/min (Ghaffarizadeh et al. 2016), and we set *U*_{i,oxy} = 20 min^{-1} (to give an oxygen diffusion length scale of about 70 μm, with steeper gradients than our typical 100 μm length scale). We set *λ*_{oxy} = 0.01 min^{-1} for a 1 mm diffusion length scale in fluid.

We set *D _{c}* = 300 μm

^{2}/min, and

*U*

_{c}= 7.2×10

^{-3}min

^{-1}(

*D*from Weinberg et al. (2007), and

_{c}*U*twice as large as the reference value in Weinberg et al. (2007) to get a smaller diffusion length scale of about 204 μm). We set

_{i,c}*λ*= 3.6×10

_{c}^{-5}min

^{-1}to give a drug diffusion length scale of about 2.9 mm in fluid.

We use *T _{D}* = 8.6 hours for apoptotic cells, and

*T*= 60 days for necrotic cells (Macklin et al., 2013). However, note that necrotic and apoptotic cells lose volume quickly, so one may want to revise those time scales to match the point where a cell loses 90% of its volume.

_{D}#### Functional forms for the birth and death rates

We model pharmacodynamics with an area-under-the-curve (AUC) type formulation. If *c*(*t*) is the drug concentration at any cell *i*‘s location at time *t*, then let its integrated exposure *E _{i}*(

*t*) be

\[ E_i(t) = \int_0^t c(s) \: ds \]

and we model its response with a Hill function

\[ R_i(t) = \frac{ E_i^h(t) }{ \alpha_i^h + E_i^h(t) }, \]

where *h *is the drug’s Hill exponent for the cell line, and *α *is the exposure for a half-maximum effect.

We model the microenvironment-dependent birth rate by:

\[ b_i(t) = \left\{ \begin{array}{lr} b_{i,P} \left( 1 – \eta_i R_i(t) \right) & \textrm{ if } \textrm{pO}_{2,P} < \textrm{pO}_2 \\ \\ b_{i,P} \left( \frac{\textrm{pO}_{2}-\textrm{pO}_{2,N}}{\textrm{pO}_{2,P}-\textrm{pO}_{2,N}}\right) \Bigl( 1 – \eta_i R_i(t) \Bigr) & \textrm{ if } \textrm{pO}_{2,N} < \textrm{pO}_2 \le \textrm{pO}_{2,P} \\ \\ 0 & \textrm{ if } \textrm{pO}_2 \le \textrm{pO}_{2,N}\end{array} \right. \]

where pO_{2,P} is the physioxic oxygen value (38 mmHg), and pO_{2,N} is a necrotic threshold (we use 5 mmHg), and 0 < *η* < 1 the drug’s birth inhibition. (A fully cytostatic drug has *η* = 1.)

We model the microenvironment-dependent apoptosis rate by:

\[ d_{i,A}(t) = d_{i,A}^* + \Bigl( d_{i,A}^\textrm{max} – d_{i,A}^* \Bigr) R_i(t) \]

*d*

_{i,A}^{max}is the maximum apoptotic death rate. We model the microenvironment-dependent necrosis rate by:

\[ d_{i,N}(t) = \left\{ \begin{array}{lr} 0 & \textrm{ if } \textrm{pO}_{2,N} < \textrm{pO}_{2} \\ \\ d_{i,N}^* & \textrm{ if } \textrm{pO}_{2} \le \textrm{pO}_{2,N} \end{array}\right. \]

*d*

_{i,N}^{*}.

##### (Illustrative) parameter values

We use *b _{i,P}* = 0.05 hour

^{-1}(for a 20 hour cell cycle in physioxic conditions),

*d*

_{i,A}^{*}= 0.01

*b*, and

_{i,P}*d*

_{i,N}^{*}= 0.04 hour

^{-1}(so necrotic cells survive around 25 hours in low oxygen conditions).

We set *α* = 30 μM*hour (so that cells reach half max response after 6 hours’ exposure at a maximum concentration *c* = 5 μM), *h* = 2 (for a smooth effect), *η* = 0.25 (so that the drug is partly cytostatic), and *d _{i,A}*

^{max}= 0.1 hour^-1 (so that cells survive about 10 hours after reaching maximum response).

### Building the Cellular Automaton Model in BioFVM

BioFVM already includes Basic_Agents for cell-based substrate sources and sinks. We can extend these basic agents into full-fledged automata, and then arrange them in a lattice to create a full cellular automata model. Let’s sketch that out now.

#### Extending Basic_Agents to Automata

The main idea here is to define an Automaton class which extends (and therefore includes) the Basic_Agent class. This will give each Automaton full access to the microenvironment defined in BioFVM, including the ability to secrete and uptake substrates. We also make sure each Automaton “knows” which microenvironment it lives in (contains a pointer pMicroenvironment), and “knows” where it lives in the cellular automaton lattice. (More on that in the following paragraphs.)

So, as a schematic (just sketching out the most important members of the class):

class Standard_Data; // define per-cell biological data, such as phenotype, // cell cycle status, etc.. class Custom_Data; // user-defined custom data, specific to a model. class Automaton : public Basic_Agent { private: Microenvironment* pMicroenvironment; CA_Mesh* pCA_mesh; int voxel_index; protected: public: // neighbor connectivity information std::vector<Automaton*> neighbors; std::vector<double> neighbor_weights; Standard_Data standard_data; void (*current_state_rule)( Automaton& A , double ); Automaton(); void copy_parameters( Standard_Data& SD ); void overwrite_from_automaton( Automaton& A ); void set_cellular_automaton_mesh( CA_Mesh* pMesh ); CA_Mesh* get_cellular_automaton_mesh( void ) const; void set_voxel_index( int ); int get_voxel_index( void ) const; void set_microenvironment( Microenvironment* pME ); Microenvironment* get_microenvironment( void ); // standard state changes bool attempt_division( void ); void become_apoptotic( void ); void become_necrotic( void ); void perform_lysis( void ); // things the user needs to define Custom_Data custom_data; // use this rule to add custom logic void (*custom_rule)( Automaton& A , double); };

So, the Automaton class includes everything in the Basic_Agent class, some Standard_Data (things like the cell state and phenotype, and per-cell settings), (user-defined) Custom_Data, basic cell behaviors like attempting division into an empty neighbor lattice site, and user-defined custom logic that can be applied to any automaton. To avoid lots of switch/case and if/then logic, each Automaton has a function pointer for its current activity (current_state_rule), which can be overwritten any time.

Each Automaton also has a list of neighbor Automata (their memory addresses), and weights for each of these neighbors. Thus, you can distance-weight the neighbors (so that corner elements are farther away), and very generalized neighbor models are possible (e.g., all lattice sites within a certain distance). When updating a cellular automaton model, such as to kill a cell, divide it, or move it, you leave the neighbor information alone, and copy/edit the information (standard_data, custom_data, current_state_rule, custom_rule). In many ways, an Automaton is just a bucket with a cell’s information in it.

Note that each Automaton also “knows” where it lives (pMicroenvironment and voxel_index), and knows what CA_Mesh it is attached to (more below).

#### Connecting Automata into a Lattice

An automaton by itself is lost in the world–it needs to link up into a lattice organization. Here’s where we define a CA_Mesh class, to hold the entire collection of Automata, setup functions (to match to the microenvironment), and two fundamental operations at the mesh level: copying automata (for cell division), and swapping them (for motility). We have provided two functions to accomplish these tasks, while automatically keeping the indexing and BioFVM functions correctly in sync. Here’s what it looks like:

class CA_Mesh{ private: Microenvironment* pMicroenvironment; Cartesian_Mesh* pMesh; std::vector<Automaton> automata; std::vector<int> iteration_order; protected: public: CA_Mesh(); // setup to match a supplied microenvironment void setup( Microenvironment& M ); // setup to match the default microenvironment void setup( void ); int number_of_automata( void ) const; void randomize_iteration_order( void ); void swap_automata( int i, int j ); void overwrite_automaton( int source_i, int destination_i ); // return the automaton sitting in the ith lattice site Automaton& operator[]( int i ); // go through all nodes according to random shuffled order void update_automata( double dt ); };

So, the CA_Mesh has a vector of Automata (which are never themselves moved), pointers to the microenvironment and its mesh, and a vector of automata indices that gives the iteration order (so that we can sample the automata in a random order). You can easily access an automaton with operator[], and copy the data from one Automaton to another with overwrite_automaton() (e.g, for cell division), and swap two Automata’s data (e.g., for cell migration) with swap_automata(). Finally, calling update_automata(dt) iterates through all the automata according to iteration_order, calls their current_state_rules and custom_rules, and advances the automata by dt.

#### Interfacing Automata with the BioFVM Microenvironment

The setup function ensures that the CA_Mesh is the same size as the Microenvironment.mesh, with same indexing, and that all automata have the correct volume, and dimension of uptake/secretion rates and parameters. If you declare and set up the Microenvironment first, all this is take care of just by declaring a CA_Mesh, as it seeks out the default microenvironment and sizes itself accordingly:

// declare a microenvironment Microenvironment M; // do things to set it up -- see prior tutorials // declare a Cellular_Automaton_Mesh CA_Mesh CA_model; // it's already good to go, initialized to empty automata: CA_model.display();

If you for some reason declare the CA_Mesh fist, you can set it up against the microenvironment:

// declare a CA_Mesh CA_Mesh CA_model; // declare a microenvironment Microenvironment M; // do things to set it up -- see prior tutorials // initialize the CA_Mesh to match the microenvironment CA_model.setup( M ); // it's already good to go, initialized to empty automata: CA_model.display();

Because each Automaton is in the microenvironment and inherits functions from Basic_Agent, it can secrete or uptake. For example, we can use functions like this one:

void set_uptake( Automaton& A, std::vector<double>& uptake_rates ) { extern double BioFVM_CA_diffusion_dt; // update the uptake_rates in the standard_data A.standard_data.uptake_rates = uptake_rates; // now, transfer them to the underlying Basic_Agent *(A.uptake_rates) = A.standard_data.uptake_rates; // and make sure the internal constants are self-consistent A.set_internal_uptake_constants( BioFVM_CA_diffusion_dt ); }

A function acting on an automaton can sample the microenvironment to change parameters and state. For example:

void do_nothing( Automaton& A, double dt ) { return; } void microenvironment_based_rule( Automaton& A, double dt ) { // sample the microenvironment std::vector<double> MS = (*A.get_microenvironment())( A.get_voxel_index() ); // if pO2 < 5 mmHg, set the cell to a necrotic state if( MS[0] < 5.0 ) { A.become_necrotic(); } // if drug > 5 uM, set the birth rate to zero if( MS[1] > 5 ) { A.standard_data.birth_rate = 0.0; } // set the custom rule to something else A.custom_rule = do_nothing; return; }

#### Implementing the mathematical model in this framework

We give each tumor cell a tumor_cell_rule (using this for custom_rule):

void viable_tumor_rule( Automaton& A, double dt ) { // If there's no cell here, don't bother. if( A.standard_data.state_code == BioFVM_CA_empty ) { return; } // sample the microenvironment std::vector<double> MS = (*A.get_microenvironment())( A.get_voxel_index() ); // integrate drug exposure A.standard_data.integrated_drug_exposure += ( MS[1]*dt ); A.standard_data.drug_response_function_value = pow( A.standard_data.integrated_drug_exposure, A.standard_data.drug_hill_exponent ); double temp = pow( A.standard_data.drug_half_max_drug_exposure, A.standard_data.drug_hill_exponent ); temp += A.standard_data.drug_response_function_value; A.standard_data.drug_response_function_value /= temp; // update birth rates (which themselves update probabilities) update_birth_rate( A, MS, dt ); update_apoptotic_death_rate( A, MS, dt ); update_necrotic_death_rate( A, MS, dt ); return; }

The functional tumor birth and death rates are implemented as:

void update_birth_rate( Automaton& A, std::vector<double>& MS, double dt ) { static double O2_denominator = BioFVM_CA_physioxic_O2 - BioFVM_CA_necrotic_O2; A.standard_data.birth_rate = A.standard_data.drug_response_function_value; // response A.standard_data.birth_rate *= A.standard_data.drug_max_birth_inhibition; // inhibition*response; A.standard_data.birth_rate *= -1.0; // - inhibition*response A.standard_data.birth_rate += 1.0; // 1 - inhibition*response A.standard_data.birth_rate *= viable_tumor_cell.birth_rate; // birth_rate0*(1 - inhibition*response) double temp1 = MS[0] ; // O2 temp1 -= BioFVM_CA_necrotic_O2; temp1 /= O2_denominator; A.standard_data.birth_rate *= temp1; if( A.standard_data.birth_rate < 0 ) { A.standard_data.birth_rate = 0.0; } A.standard_data.probability_of_division = A.standard_data.birth_rate; A.standard_data.probability_of_division *= dt; // dt*birth_rate*(1 - inhibition*repsonse) // linearized probability return; } void update_apoptotic_death_rate( Automaton& A, std::vector<double>& MS, double dt ) { A.standard_data.apoptotic_death_rate = A.standard_data.drug_max_death_rate; // max_rate A.standard_data.apoptotic_death_rate -= viable_tumor_cell.apoptotic_death_rate; // max_rate - background_rate A.standard_data.apoptotic_death_rate *= A.standard_data.drug_response_function_value; // (max_rate-background_rate)*response A.standard_data.apoptotic_death_rate += viable_tumor_cell.apoptotic_death_rate; // background_rate + (max_rate-background_rate)*response A.standard_data.probability_of_apoptotic_death = A.standard_data.apoptotic_death_rate; A.standard_data.probability_of_apoptotic_death *= dt; // dt*( background_rate + (max_rate-background_rate)*response ) // linearized probability return; } void update_necrotic_death_rate( Automaton& A, std::vector<double>& MS, double dt ) { A.standard_data.necrotic_death_rate = 0.0; A.standard_data.probability_of_necrotic_death = 0.0; if( MS[0] > BioFVM_CA_necrotic_O2 ) { return; } A.standard_data.necrotic_death_rate = perinecrotic_tumor_cell.necrotic_death_rate; A.standard_data.probability_of_necrotic_death = A.standard_data.necrotic_death_rate; A.standard_data.probability_of_necrotic_death *= dt; // dt*necrotic_death_rate return; }

And each fluid voxel (Dirichlet nodes) is implemented as the following (to turn on therapy at 21 days):

void fluid_rule( Automaton& A, double dt ) { static double activation_time = 504; static double activation_dose = 5.0; static std::vector<double> activated_dirichlet( 2 , BioFVM_CA_physioxic_O2 ); static bool initialized = false; if( !initialized ) { activated_dirichlet[1] = activation_dose; initialized = true; } if( fabs( BioFVM_CA_elapsed_time - activation_time ) < 0.01 ) { int ind = A.get_voxel_index(); if( A.get_microenvironment()->mesh.voxels[ind].is_Dirichlet ) { A.get_microenvironment()->update_dirichlet_node( ind, activated_dirichlet ); } } }

At the start of the simulation, each non-cell automaton has its custom_rule set to fluid_rule, and each tumor cell Automaton has its custom_rule set to viable_tumor_rule. Here’s how:

void setup_cellular_automata_model( Microenvironment& M, CA_Mesh& CAM ) { // Fill in this environment double tumor_radius = 150; double tumor_radius_squared = tumor_radius * tumor_radius; std::vector<double> tumor_center( 3, 0.0 ); std::vector<double> dirichlet_value( 2 , 1.0 ); dirichlet_value[0] = 38; //physioxia dirichlet_value[1] = 0; // drug for( int i=0 ; i < M.number_of_voxels() ;i++ ) { std::vector<double> displacement( 3, 0.0 ); displacement = M.mesh.voxels[i].center; displacement -= tumor_center; double r2 = norm_squared( displacement ); if( r2 > tumor_radius_squared ) // well_mixed_fluid { M.add_dirichlet_node( i, dirichlet_value ); CAM[i].copy_parameters( well_mixed_fluid ); CAM[i].custom_rule = fluid_rule; CAM[i].current_state_rule = do_nothing; } else // tumor { CAM[i].copy_parameters( viable_tumor_cell ); CAM[i].custom_rule = viable_tumor_rule; CAM[i].current_state_rule = advance_live_state; } } }

#### Overall program loop

There are two inherent time scales in this problem: cell processes like division and death (happen on the scale of hours), and transport (happens on the order of minutes). We take advantage of this by defining two step sizes:

double BioFVM_CA_dt = 3; std::string BioFVM_CA_time_units = "hr"; double BioFVM_CA_save_interval = 12; double BioFVM_CA_max_time = 24*28; double BioFVM_CA_elapsed_time = 0.0; double BioFVM_CA_diffusion_dt = 0.05; std::string BioFVM_CA_transport_time_units = "min"; double BioFVM_CA_diffusion_max_time = 5.0;

Every time the simulation advances by BioFVM_CA_dt (on the order of hours), we run diffusion to quasi-steady state (for BioFVM_CA_diffusion_max_time, on the order of minutes), using time steps of size BioFVM_CA_diffusion time. We performed numerical stability and convergence analyses to determine 0.05 min works pretty well for regular lattice arrangements of cells, but you should always perform your own testing!

Here’s how it all looks, in a main program loop:

BioFVM_CA_elapsed_time = 0.0; double next_output_time = BioFVM_CA_elapsed_time; // next time you save data while( BioFVM_CA_elapsed_time < BioFVM_CA_max_time + 1e-10 ) { // if it's time, save the simulation if( fabs( BioFVM_CA_elapsed_time - next_output_time ) < BioFVM_CA_dt/2.0 ) { std::cout << "simulation time: " << BioFVM_CA_elapsed_time << " " << BioFVM_CA_time_units << " (" << BioFVM_CA_max_time << " " << BioFVM_CA_time_units << " max)" << std::endl; char* filename; filename = new char [1024]; sprintf( filename, "output_%6f" , next_output_time ); save_BioFVM_cellular_automata_to_MultiCellDS_xml_pugi( filename , M , CA_model , BioFVM_CA_elapsed_time ); cell_counts( CA_model ); delete [] filename; next_output_time += BioFVM_CA_save_interval; } // do the cellular automaton step CA_model.update_automata( BioFVM_CA_dt ); BioFVM_CA_elapsed_time += BioFVM_CA_dt; // simulate biotransport to quasi-steady state double t_diffusion = 0.0; while( t_diffusion < BioFVM_CA_diffusion_max_time + 1e-10 ) { M.simulate_diffusion_decay( BioFVM_CA_diffusion_dt ); M.simulate_cell_sources_and_sinks( BioFVM_CA_diffusion_dt ); t_diffusion += BioFVM_CA_diffusion_dt; } }

### Getting and Running the Code

**Start a project:**Create a new directory for your project (I’d recommend “BioFVM_CA_tumor”), and enter the directory. Place a copy of BioFVM (the zip file) into your directory. Unzip BioFVM, and copy BioFVM*.h, BioFVM*.cpp, and pugixml* files into that directory.**Download the demo source code:**Download the source code for this tutorial: BioFVM_CA_Example_1, version 1.0.0 or later. Unzip its contents into your project directory.**Go ahead and overwrite the Makefile**.**Edit the makefile (if needed):**Note that if you are using OSX, you’ll probably need to change from “g++” to your installed compiler. See these tutorials.**Test the code:**Go to a command line (see previous tutorials), and test:make ./BioFVM_CA_Example_1

(If you’re on windows, run BioFVM_CA_Example_1.exe.)

### Simulation Result

If you run the code to completion, you will simulate 3 weeks of *in vitro * growth, followed by a bolus “injection” of drug. The code will simulate one one additional week under the drug. (This should take 5-10 minutes, including full simulation saves every 12 hours.)

In matlab, you can load a saved dataset and check the minimum oxygenation value like this:

MCDS = read_MultiCellDS_xml( 'output_504.000000.xml' ); min(min(min( MCDS.continuum_variables(1).data )))

And then you can start visualizing like this:

contourf( MCDS.mesh.X_coordinates , MCDS.mesh.Y_coordinates , ... MCDS.continuum_variables(1).data(:,:,33)' ) ; axis image; colorbar xlabel('x (\mum)' , 'fontsize' , 12 ); ylabel( 'y (\mum)' , 'fontsize', 12 ); set(gca, 'fontsize', 12 ); title('Oxygenation (mmHg) at z = 0 \mum', 'fontsize', 14 ); print('-dpng', 'Tumor_o2_3_weeks.png' ); plot_cellular_automata( MCDS , 'Tumor spheroid at 3 weeks');

#### Simulation plots

Here are some plots, showing (left from right) pO_{2} concentration, a cross-section of the tumor (red = live cells, green = apoptotic, and blue = necrotic), and the drug concentration (after start of therapy):

##### 1 week:

Oxygen- and space-limited growth are restricted to the outer boundary of the tumor spheroid.

##### 2 weeks:

Oxygenation is dipped below 5 mmHg in the center, leading to necrosis.

##### 3 weeks:

As the tumor grows, the hypoxic gradient increases, and the necrotic core grows. The code turns on a constant 5 micromolar dose of doxorubicin at this point

##### Treatment + 12 hours:

The drug has started to penetrate the tumor, triggering apoptotic death towards the outer periphery where exposure has been greatest.

##### Treatment + 24 hours:

The drug profile hasn’t changed much, but the interior cells have now had greater exposure to drug, and hence greater response. Now apoptosis is observed throughout the non-necrotic tumor. The tumor has decreased in volume somewhat.

##### Treatment + 36 hours:

The non-necrotic tumor is now substantially apoptotic. We would require some pharamcokinetic effects (e.g., drug clearance, inactivation, or removal) to avoid the inevitable, presences of a pre-existing resistant strain, or emergence of resistance.

##### Treatment + 48 hours:

By now, almost all cells are apoptotic.

##### Treatment + 60 hours:

The non-necrotic tumor is nearly completed eliminated, leaving a leftover core of previously-necrotic cells (which did not change state in response to the drug–they were already dead!)

### Source files

You can download completed source for this example here: https://sourceforge.net/projects/biofvm/files/Tutorials/Cellular_Automaton_1/

This file will include the following:

- BioFVM_cellular_automata.h
- BioFVM_cellular_automata.cpp
- BioFVM_CA_example_1.cpp
- read_MultiCellDS_xml.m (updated)
- plot_cellular_automata.m
- Makefile

### What’s next

I plan to update this source code with extra cell motility, and potentially more realistic parameter values. Also, I plan to more formally separate out the example from the generic cell capabilities, so that this source code can work as a *bona fide *cellular automaton framework.

More immediately, my next tutorial will use the reverse strategy: start with an existing cellular automaton model, and integrate BioFVM capabilities.

Return to News • Return to MathCancer • Follow @MathCancer

## Paul Macklin calls for common standards in cancer modeling

At a recent NCI-organized mini-symposium on big data in cancer, Paul Macklin called for new data standards in Multicellular data in simulations, experiments, and clinical science. USC featured the talk (abstract here) and the work at news.usc.edu.

**Read the article:** http://news.usc.edu/59091/usc-researcher-calls-for-common-standards-in-cancer-modeling/ (Feb. 21, 2014)

## Macklin Lab featured on March 2013 cover of Notices of the American Mathematical Society

I’m very excited to be featured on this month’s cover of the Notices of the American Mathematical Society. The cover shows a series of images from a multiscale simulation of a tumor growing in the brain, made with John Lowengrub while I was a Ph.D. student at UC Irvine. (See Frieboes et al. 2007, Macklin et al. 2009, and Macklin and Lowengrub 2008.) The “about the cover” write-up (Page 325) gives more detail.

The inside has a short interview on our more current work, particularly 3-D agent-based modeling. You should also read Rick Durrett‘s perspective piece on cancer modeling (Page 304)—it’s a great read! (And yup, Figure 3 is from our patient-calibrated breast cancer modeling in Macklin et al. 2012. 😉 )

The entire March 2013 issue can be accessed for free at the AMS Notices website:

http://www.ams.org/notices/201303/

I want to thank Bill Casselman and Rick Durrett for making this possible. I had a lot of fun in the process, and I’m grateful for the opportunity to trade ideas!

## DCIS modeling paper accepted

I am pleased to report that our paper has now been accepted. You can download the accepted preprint here. We also have a lot of supplementary material, including simulation movies, simulation datasets (for 0, 15, 30, adn 45 days of growth), and open source C++ code for postprocessing and visualization.

I discussed the results in detail here, but here’s the short version:

- We use a mechanistic, agent-based model of individual cancer cells growing in a duct. Cells are moved by adhesive and repulsive forces exchanged with other cells and the basement membrane. Cell phenotype is controlled by stochastic processes.
- We constrained all parameter expected to be relatively independent of patients by a careful analysis of the experimental biological and clinical literature.
- We developed the very first patient-specific calibration method, using clinically-accessible pathology. This is a key point in future patient-tailored predictions and surgical/therapeutic planning.
- The model made numerous quantitative predictions, such as:
- The tumor grows at a constant rate, between 7 to 10 mm/year. This is right in the middle of the range reported in the clinic.
- The tumor’s size in mammgraphy is linearly correlated with the post-surgical pathology size. When we linearly extrapolate our correlation
*across two orders of magnitude*, it goes right through the middle of a cluster of 87 clinical data points. - The tumor necrotic core has an age structuring: with oldest, calcified material in the center, and newest, most intact necrotic cells at the outer edge.
- The appearance of a “typical” DCIS duct cross-section varies with distance from the leading edge; all types of cross-sections predicted by our model are observed in patient pathology.
- The model also gave new insight on the underlying biology of breast cancer, such as:
- The split between the viable rim and necrotic core (observed almost universally in pathology) is not just an artifact, but an actual biomechanical effect from fast necrotic cell lysis.
- The constant rate of tumor growth arises from the biomechanical stress relief provided by lysing necrotic cells. This points to the critical role of intracellular and intra-tumoral water transport in determining the qualitative and quantitative behavior of tumors.
- Pyknosis (nuclear degradation in necrotic cells), must occur at a time scale between that of cell lysis (on the order of hours) and cell calcification (on the order of weeks).
- The current model cannot explain the full spectrum of calcification types; other biophysics, such as degradation over a long, 1-2 month time scale, must be at play.

**Direct link to the preprint:**http://www.mathcancer.org/Publications.php#macklin12_jtb

## PSOC Short Course on Multidisciplinary Cancer Modeling

Next Monday (October 17, 2011), the USC-led Physical Sciences Oncology Center / CAMM will host a short course on multidisciplinary cancer modeling, combining the expertise of biologists, oncologists, and physical scientists. I’ll attach a PDF flyer of the schedule below. I am giving a talk during “Session II – The Physicist Perspective on Cancer.” I will focus on tailoring mathematical models from the ground up to clinical data from individual patients, with an emphasis on using computational models to make testable clinical predictions, and using these models a platforms to generate hypotheses on cancer biology.

The response to our short course has been overwhelming (in a good way), with around 200 registrants! So, registration is unfortunately closed at this time. However, the talks will be broadcast live via a webcast. The link and login details are in the PDF below. I hope to see you there! — Paul

**Agenda**

7:00 am – 8:25 am : **Registration, breakfast, and opening comments, etc.**

David B. Agus, M.D. (Director of USC CAMM)

W. Daniel Hillis, Ph.D. (PI of USC PSOC, Applied Minds)

Larry A. Nagahara (NCI PSOC Program Director)

8:30 am – 10:15 am : **Session I – Cancer Biology and the Cancer Genome**

Paul Mischel, UCLA – The Biology of Cancer from Cell to Patient, Oncogenesis to Therapeutic Response

Matteo Pellegrini, UCLA – Evolution in Cancer

Mitchelll Gross, USC – Historical Perspective on Cancer Diagnosis and Treatment

10:30 am – 12:15 pm : **Session II – The Physicist Perspective on Cancer**

Dan Ruderman, USC – Cancer as a Multi-scale Problem

Paul Macklin, USC – Computational Models of Cancer Growth

Tom Tombrello, Cal-Tech – Perspective: Big Problems in Physics vs. Cancer

1:45 pm. – 3:10 pm : **Session III – Novel Measurement Platforms & Data Management & Integration**

Michelle Povinelli, USC – The Role of Novel Microdevices in Dissecting Cellular Phenomena

Carl Kesselman, USC – Data Management & Integration Challenges in Interdisciplinary Studies

3:10 pm – 3:40 pm : **Session IV – Creativity in Research at the Interface between the Life and Physical Sciences**

‘Fireside Chat’ David Agus and Danny Hillis, USC

4:00 pm – 5:00 pm : **Capstone – Keynote Speaker**

Tim Walsh, Game Inventor, Keynote Speaker

5:30 pm – 8:00 pm : **Poster Session and Reception**

**Click here for the poster, including webcast information: http://mathcancer.org/publications/PSOC_short_course_2011.pdf**

## DCIS paper resubmitted; lots of clinical predictions, lots of validation

After a lot of revision, I have merged my two papers on ductal carcinoma in situ (DCIS) in to a single manuscript and resubmitted to the Journal of Theoretical Biology for final review. A preprint is posted at my website, along with considerable supplementary material (data sets and source code, and animations). Thanks to my co-authors and other friends for all the help in the revisions. Thanks also the reviewers for insightful comments. I think this revised manuscript is all the better for it.

DCIS is a precursor to invasive breast cancer, and it is generally detected by annual mammographic screening. More advanced DCIS (with greater risk) tends to have comedonecrosis–a type of cell death that leaves calcium phosphate deposits in the centers of the ducts. In fact, this is generally what’s detected in mammograms. DCIS is usually surgically removed by cutting out a small ball of tissue around what’s found in the mammograms (breast-conserving surgery, or lumpectomy). But current planning isn’t so great. Even with the state-of-the-art in patient imaging and surgical planning, about 20%-50% of women need to get a second surgery because the first one didn’t get the entire tumor.

So, there’s great need to understand calcifications, and how what you see in mammography relates to the actual tumor size and shape. And if you do have a model to do this, there’s great need to calibrate it to patient pathology data (the stuff you get from your biopsies) so that the models say something meaningful about individual patients. And there has been no method to do that. Until now.

(As far as I know), this paper is the first to calibrate to individual patient immunohistochemistry and histopathology. This, along with some parameter estimates to the theoretical and experimental biology literature, allows us to fully constrain the model. No free parameters to play with until it looks right. Any results are fully emergent from a mechanistic model and realistic parameter estimates rooted in the biology.

This model also includes the most detailed description of necrosis–the type of cell death that results in the comedonecrosis seen in mammograms. We include cell swelling, cell bursting, gradual loss of fluid, and the very first model of calcification.**Clinical predictions, with lots of validation:**

All said and done, the model gives some big (and validated!) predictions:

- The model predicts that a tumor grows through the duct at a constant rate. This is consistent with what’s actually seen in mammography.
- The model gives a new explanation for the known trend: when necrotic cells burst and lose fluid, it makes it more mechanically favorable for proliferating cells to push into the center of the duct, rather than along the duct. For this reason the model predicts faster growth in smaller ducts, and slower growth in larger ducts.
- The model predicts growth rates between 7.5 and 10 mm per year. This is quantitatively consistent with published values in the clinical literature.
- The model predicts the difference between the size in a mammogram and the actual size (as measured by a pathologist after surgery) grows in time. This unfortunately means that it’s unlikely that there is some “fixed” safe distance to cut around the mammographic findings.
- On the other hand, the model predicts that there is a linear correlation between the size in a mammogram and the actual (pathology) tumor size. This bodes well for future surgical planning.
- Better still, the linear correlation we found quantitatively fits through 87 published patients, spanning
*two orders of magnitude*.

**New insights on DCIS biology:**

The model also makes several key predictions on the smaller-scale biology:

- The model predicts that fast swelling of necrotic cells (on the order of 6 hours) is responsible for the tear between the viable rim and necrotic core seen in just about every pathology image of DCIS.
- The model predicts that the necrotic core is “age structured”, with newly necrotic cells (with relatively intact nuclei) on the outer edge, and interior band of mostly degraded but noncalcified cells, and a central core of oldest, calcified material. This compares well with patient histopathology.
- Comparing the model-predicted age structuring to histopathology predicts a sharper estimate on the various necrosis time scales: swelling and lysis (~6 h) < slow fluid loss (~days to a week) < pyknosis (~10+ days) < calcification (~2 weeks).
- Because the model only predicts linear / casting-type calcifications (long “plugs” of calcification), other biophysics must be responsible for the variety of calcification types seen in mammography.
- Among other mechanisms, we postulate a very long-timescale (1-2 months) process of degradation of the phospholipid “backbone” of the calcifications, resulting in degradation of the calcifications. The cracks seen in the central portions of calcifications (in histopathology) supports this view.

*spontaneous resolution:*where calcifications sometimes disappear from mammograms, while the underlying tumor is still present.

**Long-term outlook, and next steps:**