ConsortiaFlux

ConsortiaFlux is a web app for

• visualizing
• modifying
• analyzing

genome scale metabolic models of microbial consortia.

ConsortiaFlux is powered by an Express server which communicates with a MongoDB database and runs Python scripts using cobrapy (COnstraints Based Reconstruction and Analysis). The web app is a single page application built off of AngularJS.

Usage

See the user guide (coming soon).

Installation

The web-app and server each depend on NodeJS and npm.

Node.js

If you are on OS X or Linux system, I recommend installing using one of these methods to avoid having to sudo. Furthermore see here to set up npm packages to install into a custom global directory without sudo. You can check if the install worked by running

node -v

in a terminal; it should return the version number. You don't need to worry about sudo on Windows, as Windows runs as admin by default. However, you may face some PATH issues on Windows (see below).

npm

npm is node package manager and is the "largest ecosystem of open source libraries in the world". Node comes with npm, though it is usually not the latest version, and more importantly, it is not located where your global npm modules are installed. Before continuing, you should make sure your PATH variable catches the correct npm binary, or in simpler terms, running

npm -v
npm install -g npm
npm -v

should return a different value on the second npm -v. (You may need to open and close the terminal first). If it is not, compare the path returned by npm install -g npm (of the format path -> path2) to which npm. If they are different, it is because the directory containing Node's npm is earlier in your PATH than the one containing the binaries of npm's global modules. For example, I have the following in my .bashrc:

# Node and NPM
export NPM_PACKAGES="$HOME/node/npm-packages"
export NODE_PATH="$NPM_PACKAGES/lib/node_modules:$NODE_PATH"
export PATH="$PATH:$NPM_PACKAGES/bin:$HOME/node/bin"
export MANPATH="$NPM_PACKAGES/share/man:$MANPATH"

You can see PATH is the original PATH, followed by npm-packages, followed by node. On OS X and Linux, you can add the above to your .bashrc or .profile, etc. to achieve the same effect. To see your environment's PATH, run echo $PATH. Note, I used the two methods mentioned above for not using sudo with Node and npm, with custom folder names (node, and npm-packages). On Windows, you can go to System -> Advanced Settings -> Environment Variables -> Path and edit it there.

Web app

Once Node and npm are all properly set up, installation and local hosting of the web app should be a breeze. First, lets install bower and gulp as global npm modules:

npm install -g bower gulp

Once we have those, cd into the web-app folder and run

bower install
npm install

These commands will read through the bower.json and package.json and install dependencies into bower_components and node_modules, respectably. When you are ready to view the application, simply run

gulp

and your browser will open to http://localhost:3000. At this point, any source modifications will immediately trigger a page refresh.

MongoDB

See the downloads or install instructions. On OS X there a few things you need to do before starting mongod, see Run MongoDB. Alternatively, Run page for Ubuntu.

Express Server

Same as above, except no bower here, and there may be some annoying Python dependencies that refuse to build when installed through pip.

cd express-server
npm install -g nodemon
npm install

There are some scripts you can run to curl requests to the localserver (optimize will only work if cobrapy is installed fully):

./build-models.sh
curl http://localhost:9001//model/optimize/iJO1366

cobrapy

See installation instructions. The optional dependencies required are:

• libsbml
• numpy
• scipy

Whether or not you use virtualenv is up to you. I tried to get libsbml working with a virtual environment, however something with the C bindings not linking properly prevented it from working. As a global install however, it worked fine. If (sudo) pip install python-libsbml is not working for you (the build fails), you should have better luck building from source; see building and installing libsbml. Once you have the libsbml source, you can run

./config --prefix=DIR/lib/python-version/dist-packages/ --with-python --with-python-interpreter=/path/to/python/binary

Also, I even had to use

--with-libxml=/path/to/old/libxml/version

even though my current libxml was multiple versions ahead of that required. SWIG is required to build bindings between C and Python for libsbml.

Please note, you will need to point to the correct Python binary (the one with a working cobra install) in line 37 of routes/model/optimize/index.js:

var optimizeScript = cp.spawn('python', args)

This will be moved into config.js soon.

Changelog (web-app)

Current: 1.5

• Renders metabolic network as a set of nodes and links
• Integration of d3 with custom canvas rendering for improved performance
• Basic dragging of nodes working
• Panning and zooming of canvas working
• (1.1.x) Added deletion and adding, beta and WIP
  • removal of nodes from target/source when trying to create a reaction is now removed from the options menu
  • (1.1.1) Error handling
  • (1.1.2) Significant performance increases
• (1.2.x) Re factored code for an MVC model, significant overhaul changes
• (1.3.x)
  • Added the specie extracellular network and added new specie class that extends node
  • (1.3.1) Fixed reaction visual bug
• (1.4.x) You can now enter a specie from a network
  • (1.4.1) Fixed broken adding/populate options. Few more modular improvments
  • (1.4.2) Made it more modular when it comes to adding new systems
  • (1.4.3) Added the ability to return to the network
  • (1.4.4) Zooming in and out of the specie/system returns to earlier state
• (1.5.x) Created brand new data structure to implmented a 3D graph structure

Community Flux Balance Analysis Pipeline

Models have varying names for metabolites, or sometimes none and only the id. Develop a pipeline for creating and maintaining a standardized set of metabolites and reactions to be used across all models. Potentially integrate this with the web application to increase cooperation and standardization.

1. Given a list of n species and their respective genome scale metabolic models (following SBML format as either originally an .xml or later as a .json),
2. Develop a list of all reactions across all species that occur in the extracellular space, or e
3. For each species i in n, loop through every reaction in e. If a reaction contains a metabolite that is involved in a reaction from the 'global e reactions', append it into the model of that specie. This should take O(rxns_e * n) where rxns_e is the maximal number of extracellular reactions in a specie, and n is the number of species.
4. Perform FBA individual on each species i in n, optimizing for biomass production. The time this takes is proportional to the size of the linear problem to be solved. Further time analysis is necessary to accurately provide an upper bound on this process.
5. Using the results of FBA on each species, we have gathered a flux for each reaction in rxns_e of that specie. This will be different across the species; set a new upper and lower bound for each of this reactions by taking the average - SD and the average + SD. In this way, for all of the reactions which share extracellular metabolites, after this point, they will have the same lower and upper bounds.
6. Perform FBA on each species again. Optimizing for biomass, as before. Again, more structured time analysis is needed, however this should take similar time to the previous step.
7. Each biomass objective function for each species returns a numeric value. Use z-scoring to standardize these distributions, so that we may compute a 'relative biomass' value for each specie. The sum of all these values must be equal to 1.
8. Construct a single metabolic model representing the entire community. Each species will be assigned a 'compartment'. The upper and lower bounds of each reaction in each specie will be modified as per that specie's relative biomass. In this way, reactions belonging to a less important specie will have less effect (or more) on the optimization of the objective function, which will be biomass. The results of this cFBA analysis should coincide (to a degree) with experimentally observed community dynamics.

License

MIT License