Friday, November 26, 2010

Society for Neuroscience meeting 2010: a Python perspective

Last week I was at the annual meeting of the Society for Neuroscience (SfN) in San Diego. For those who aren't familiar with the meeting, it is huge, with over 30000 participants: it seemed as though every flight into San Diego on the day before the meeting was filled with people carrying poster tubes. Even as far away as Paris, the check-in line for every America-bound flight from Charles de Gaulle was bristling with posters.

The first few SfN meetings I went to, I exhausted myself trying to see every poster and every talk that could possibly be of interest (while lugging round several kilograms of abstract books and free journal samples from Publishers' Row). This time I focused on the early visual system (there was a particularly good symposium on colour vision) and on neuroinformatics (in particular database and simulation software) (and Publishers' Row this year was at the far, far end of the enormous poster/exhibitors hall - I never got that far).

Anyway, this is a report on the meeting from a software (and particularly Python software) perspective.

NEURON and Python workshop

On Saturday morning, I took part in a workshop/seminar on NEURON+Python, organised by Tom McTavish, Ted Carnevale and Tom Morse from Yale. This was apparently hugely over-subscribed, which certainly reflects the large NEURON user community, but it is clear that all the participants were intrigued by the potential power that Python brings to this well-established simulator (in a related note, I was just told that the article we wrote about NEURON and Python is one of the 25 most-cited articles in the Frontiers in Neuroscience journal series).

The format was part traditional presentations, part interactive tutorial using Sage notebooks.

Tom McTavish gave an interative introduction to Python, and then to using Python with the NEURON simulator. Tom Morse talked about ModelDB, and Ted Carnevale discussed using Python and Hoc (NEURON's original scripting language) together. Tom Foutz presented Neuron3D, a Mayavi-based tool for working with 3D views of neurons, nuclei and electrodes in NEURON. This looked very nice, and reminded me of a couple of tools presented at the last CodeJam meeting, Mike Hull's morphforge and Michele Mattioni's NeuronVisio, both of which also use Mayavi. Mike was in San Diego at the meeting, so I introduced him to Tom, and they started hatching plans for some collaboration - watch this space. I talked about a miscellaneous bunch of stuff related to neuronal simulation and Python, including PyNN, Sumatra, NeuroTools, and nrnutils.

Materials from the workshop will be available soon on the NEURON forum.


A search for "python" in the Neuroscience Meeting Planner gives nine abstracts, all of which refer to the programming language (it appears no-one is reporting research into the nervous system of the family Pythonidae). ("Matlab" gives 36 abstracts, so it's clear we've a way to go). I guess many more people probably use Python in their research, but at least this gives an idea of how many thought it was important enough to mention in the abstract.

The topics covered included neuroscience education (25.27), the role of relative ion channel densities in neuronal firing properties in tadpoles (46.8), the CARMEN e-science platform for neurophysiology data (107.1), web-based model presentation using NEURON and Sage (208.27), the eagerly-anticipated 2.0 release of the NEST simulator (208.28), modelling olfactory neurons (369.6), analysis of spectroscopy data in Alzheimer's disease (746.20), inference of brain connectivity (918.13), and boosting your research productivity using NeuroDebian (918.15).

I'll say more about CARMEN and NeuroDebian below.

Although I didn't visit the poster, the abstract on modelling the olfactory bulb external tufted cell (369.6) interests me because they implemented versions of the model in both Python and Matlab. The importance of implementing neuroscience models in at least two simulation environments as a way of checking the numerical methods of a given simulator don't impact the qualitative results and of improving the reproducibility of the model is a subject I've been banging on about for some time, so it's great to see that other people are also doing this. (Great to see this cell type being modelled, incidentally. When I find time to return to olfactory bulb modelling, I'll certainly follow up on this work).

I really liked Tom McTavish's poster (208.27) on using Sage to interact with NEURON models over the web. Another subject I often go on about is enhancing the reusability of neuronal models: building new models based on well-tested and well-understood existing components, rather than building from scratch every time, is something I think there is too little of in neuroscience at the moment, so tools that improve the accessibility, comprehensibility and testability of neuroscience models are to be welcomed. There is nothing NEURON-specific about Sage, of course: I think this is a tool that would be useful for any Python-supporting neuroscience simulator.

INCF booth

The International Neuroinformatics Coordinating Facility (INCF) had a booth at SfN, at which they had a program of live demonstrations of neuroinformatics software and resources (they also had a very comfy sofa), including a demo of a Python client for XNAT-based neuroimaging databases by Yannick Schwartz and colleagues from LNAO at NeuroSpin.

Neurodebian booth

Neurodebian is "a platform that provides a staging area for neuroscience software packages on their way into Debian". They had a booth at SfN, and I swung by to discuss packaging PyNN for Debian. I now have a short list of things to do to make PyNN suitable for packaging, so hopefully the 0.7.0 release will be available in NeuroDebian. Overall, they have an impressive selection of packages. I am almost convinced to try running NeuroDebian on our cluster, instead of Ubuntu. One package I need that is missing is mpi4py, but the Neurodebian guys are now planning to add that. Here is their own report on the meeting.

Other stuff

Some non-Python stuff is definitely worth a mention. I've been following the CARMEN project for some time. This is a UK-funded project to develop a portal for sharing of neuroscience data and analyses. They have now opened their portal to academic users. They are also responsible for a number of advances and fledgling standards in electrophysiology data sharing, including the MINI specification (Minimum Information about a Neuroscience Investigation) and the NDF (Neurophysiology Data translation Format) (not to be confused with NDF (Neurospaces Description Format) :-). 

Gwen Jacobs was presenting the Yogo Data Management Framework, "a set of software tools created to rapidly build scientific data-management applications". I think this sort of tool - enabling straightforward customization of database systems for the needs of individual labs - is going to be essential in bringing some of the vast quantity of neuroscience data generated every year into a form where it can be shared, reanalyzed and reused. Although the system we're developing in my group, Helmholtz, is much more neuroscience-specific and less general than Yogo (and much less polished, at the moment), it was interesting to see that we've made many of the same architectural decisions, including building on top of a rapid-development web framework (Ruby on Rails for Yogo, Django in our case).

Finally, if you haven't already done so, check out the demo video for the Whole Brain Catalog, the software for which is open-source, and, I just saw, exposes a RESTful API.