Database Reference
In-Depth Information
So for me, the question is: Why is right now the right time to be trying to
build tools to solve these neuroscience problems? One reason is that the
data are going to be available very shortly. About five years ago there was a
real question of tracing out all the neurons in neural systems and how they
can connect, because no one had the schematic of the brain up until then.
Connectome projects are projects that are tackling this goal of creating a
comprehensive map of the neural connections of the brain. So people have
been building these connectomes of really dense schematics of the systems,
led by groups such as Sebastian Seung's group at MIT.
However, the problem with biological circuits is that the schematic that you
get out doesn't have a nice little box drawn around parts saying “This is an
adder” or “This is a register.” No, it's just this dense graph of crap. You can
imagine what it's like by trying to figure out how a processor works by just
looking at how all the transistors are connected. Obviously, that really limits
your understanding. So Konrad Kording
1
, a scientist at Northwestern, and
I started trying to build models to discover the structure and patterns in this
connectomic state. We have a paper that was just sent out for review on
exactly this idea of how—given this high-throughput, ambiguous, noisy, some-
times error-filled data—you actually extract out scientific meaning.
The analogy here to bioinformatics is really strong. It used to be that a biologist
was a biologist. And then we had the rise of genomics as a field, and now you
have computational genomics as a field. The entire field of bioinformatics
is actually a field where people who are biologists just sit at a computer.
They don't actually touch a wet lab. It became a real independent field
partially because of this transition toward the availability of high-quality,
high-throughput data. I think neuroscience is going to see a similar transition.
I like to say that neuroscience is generally ten to fifteen years behind the rest
of biology because, in many ways, it's a harder problem: there's more ambigu-
ity, and getting the data is much, much harder. So the hope is that right now is
the right time to strike.
Scott Linderman
2
, at Harvard, and I are organizing a workshop at the 2014
Computational and Systems Neuroscience [COSYNE] conference on discov-
ering structure in neural data, organized around questions like: How do we
find these items? And how do we build algorithms to find patterns in this data?
In ten years, the data's going to be there, and if people just keep taking the
Fourier transform or keep doing PCA on this data, then we're really going to
be screwed. There's just no way you're going to understand these systems.
