parWholeCell

Project to develop a parallelized whole-cell model simulator.

Currently implements a simplified whole-cell model using object-oriented MATLAB.

As of 3/12/2013 the code implements a simple whole-cell model containing 7 submodels. The submodels are extremely simple.

• Complexation
• Metabolism: flux-balance analysis
• RNA degradation
• RNA maturation: processing, cleavage, etc.
• Protein maturation: maturation, translocation
• Transcription
• Translation

and 3 states:

• MoleculeCounts: metabolite, RNA, protein monomer, complexes. Each RNA and protein has two forms: nascent and mature.
• Metabolism: growth and reaction fluxes
• Mass

The model includes 3 compartments:

• Cytosol
• Membrane
• Extracellular space

Requirements

• MATLAB >= R2012b
• MATLAB toolboxes
  • bioinformatics

Installation

1. Clone from github repository
   • https://github.com/CovertLab/parWholeCell.git
   • ssh://[email protected]:CovertLab/parWholeCell.git
2. Open MATLAB, change to parWholeCell directory
3. In addition, setup the MATLAB warnings and path at the beginning of each MATLAB session

   setWarnings();
   setPath();

4. Thats it!

How to run simulation and plot

Use the code below to run a simulation with the default options and parameter values.

%set warnings and path (need to execute once per session)
setWarnings();
setPath();

%simulate and plot
runSimulation(); %example use in wholecell.sim.SimulationTest.testRunSimulation

How to run tests

parWholeCell uses MATLAB XUnit to run unit tests. Use the code below to build fixtures and run tests.

%set warnings and path (need to execute once per session)
setWarnings();
setPath();

%generate fixtures for tests so tests run faster
generateTestFixtures();

%Run tests. Tests are logged to out/test/results.xml in JUnit-style XML
runTests();

Simulation API

import wholecell.sim.Simulation;

sim = Simulation()

%get options, parameters
options = sim.getOptions();
parameters = sim.getParameters();

%set options, parameters
options = struct('lengthSec', 10);
sim.setOptions(options);
sim.setParameters(parameters);

%get states, process
s = sim.getState(<stateId>)
p = sim.getProcess(<processId>)

%run simulation
loggers = {wholecell.sim.logger.Shell()};
sim.run(<optional cell array of loggers>);

%calculate initial conditions
sim.calcInitialConditions();

%calculate one time step
sim.evolveState()

Comparison to Karr et al., 2012 version

The main ideas of this model are identical to the 2012 version. The models differs in a few ways:

• The Metabolite, RNA, ProteinMonomer, and ProteinComplex states have been merged into one state called MoleculeCounts
• The RNA and protein maturation submodels have been merged into 2 submodels
• The model includes only three compartments. The terminal organelle and nucleoid compartments have been eliminated
• As of 3/12/2013 there are no submodels to represent DNA replication, cytokinesis, or protein decay
• The submodels are extremely simple
• There are only 2 RNA and protein forms: nascent and mature

Additionally, the implementation differs in a few ways:

• There is no database implementation of the knowledge base. Instead one class wholecell.kb.KnowledgeBase provides the same functionality as before. This class reads data from a Excel workbook.

• The names of the submodels, states, and methods, and properties have been modified in a few places for clarity.

• Most importantly, the process-state communication has been improved, keeping in mind whats needed to parallelize the implementation. We've introduced a new concept called a partition which represents the interface between a process and a state. The partitions do several things.

  • They encapsulate the mapping between states and proceses.
  • They also isolate the submodels such that they don't read/write the same piece of memory.

  States have a list of partitions and several methods for managing them.

  • addPartition: called during model construction to create a new process-state mapping
  • partition: splits a state in substates/partitions assigned to each process
  • merge: merges substates back into the full state

• It is now possible to vary the time step size

• There is no master table of parameters yet. This is easy to implement. I just haven't done it yet.

• There's no code for model fitting or analysis.

• The disk logger has been simplified. The logging functionality is the same. But there no reindexing and less support for efficient retrieval.

Credits

parWholeCell was developed by researchers at Stanford University and Intel:

• Markus Covert Systems Biology Lab, Stanford University
  • Markus Covert. Contact [email protected].
  • Jonathan Karr. Contact [email protected].
• Parallel Computing Lab, Intel Corporation
  • Sanchit Misra. Contact [email protected].
  • Kiran Pamnany. Contact [email protected].
  • Mikhail Smelyanskiy. Contact [email protected]