django-broab

Django package that implements the a set of models and an API for neurophysiological data using the Neo model.

Key features

• takes advantage of Postgres' support for Arrays (e.g. for AnalogSignal.signal and SpikeTrain.waveforms)
• takes advantage of Postgres' support for Hstore key-value pairs (for annotations)
• models can be inherited to let developers define custom data models with additional fields or relationships while still conforming to the Neo data model (for example, see github.com/gentnerlab/sturnus)
• to maximize interoperability with Python, Matlab, & R, data is exposed via a RESTful API

"Quick" start

0. If you aren't already running a postgres backend...

   1. Create a new database, then install the hstore extension:

      CREATE EXTENSION hstore;
      

   2. Install psycopg2 to interface with your Postgres database:

      pip install psycopg2
      

   3. Configure your Postgres database in settings.py:

      DATABASES = {
          'default': {
              'ENGINE': 'django.db.backends.postgresql_psycopg2',
              ...
              }
          }
      

1. Install the dependencies:

   pip install django-model-utils
   pip install djorm-ext-pgarray
   pip install djorm-ext-hstore
   

4. Add "broab" to your INSTALLED_APPS setting like this:

   INSTALLED_APPS = (
       ...
       'broab',
   )
   

5. Run python manage.py syncdb to create the models.

3. Start the development server and visit http://127.0.0.1:8000/admin/ to create broab objects (you'll need the Admin app enabled).
4. Visit http://127.0.0.1:8000/admin/broab/ to view objects available.

Notable deviations from Neo

A number of data models in Neo are simply "array" versions of unidimensional data structures, e.g. AnalogSignal/AnalogSignalArray, Spike/SpikeTrain, Event/EventArray, Epoch/EpochArray. In order to simplify the schema and take advantage of a relational database implementation of Neo, there are some key differences between broab and Neo.

1. The "Array" form of most objects does not exist in Broab. While arrays are more efficient for analysis, they make it more difficult to query within the array. (However, I've considered adding Array models as targets of ForeignKeys from their respective data model to group Events & AnalogSignals)
2. The exception to this is SpikeTrain, which is maintained while Spike is not. This is because it is rare that one wants to filter a query of Spikes in a SpikeTrain according to some Spike-specific feature. Maintaining only SpikeTrains in the database should be sufficient for the vast majority of queries while also keeping the size of the SpikeTrain table down by an order of magnitude.
3. Epochs do not exist in Broab. Rather, Events have an optional "duration" field. This is consistent with proposed changes to Neo.

Future goals

1. Write a neo.io.BroabIO for a clean export/import to and from any Neo-supported file format