Print Friendly
E-mail Page
Share Page
One on One with Terry Sejnowski
Recently elected to the Institute of Medicine of the National Academies, Terry Sejnowski is credited with pioneering the field of computational neuroscience in the 1980s by drawing on his training in both physics and neurobiology. His laboratory, staffed by an eclectic team with backgrounds in mathematics, electrical engineering, physics, and even philosophy, uses computer models based on physiological experiments to make predictions that call for additional experiments to test the model. Along the way, his lab has contributed major breakthroughs in understanding how the brain works, developed powerful technology that is used in industry, and helped launched many successful scientific careers.
How did you become interested in the brain and eventually decide to apply your training in physics to Neuroscience?
After I finished my Ph.D. doing a computer-modeling thesis about neurons,
I took a course at Woods Hole in Neurobiology where I learned a lot of
really exciting techniques that taught me how neuroscientists collect and
analyze data. I realized as a consequence of that course that what I really
would like to do for my career was to integrate the two. There were other
people who had been doing theoretical work before me, like John Hopfield,
who was my mentor at Princeton. And there were many people doing
physical experiments, but there were very few who were capable of being
able to both think of a good theoretical modeling study and then design
a good experiment to carry it out and do that in the same lab. So that is
really what makes my lab almost unique.
You pioneered the field of Computational Neuroscience. Can you put in perspective how this approach has changed scientific research?
There's an enormous shift that's occurred in the last 10 years in
neuroscience as new techniques have made it possible to collect larger
and more complex data sets, and the work we've done in computational
neuroscience has helped to analyze and interpret these data. The new
optical recording techniques that we're now developing require even higher
throughput computer systems that wouldn't have been possible 10 years
ago because computers weren't fast enough. This new technology allows
us to make progress much faster, and to test the theoretical modeling
studies and make more powerful predictions.
One of the areas of interest in your lab is how the brain creates and stores memory. How much do we know at this point about that process in the brain?
I don't think we've gotten to the bottom of it yet. We know where to look
and we have some tantalizing hints from the biochemical and the physiological
studies that have been done, primarily in the hippocampus and
the cerebral cortex. However, the link between these changes that occur
at synapses and changes in behavior is very weak. There are hundreds of
microscopic changes that occur in neurons, so how do you know which
one of those is going to be relevant for explaining the changes in behavior?
One of the advantages now is a highly detailed synaptic modeling
program we've developed called MCell, which is giving us a very powerful
way to put together all the information that neuroscientists have found at
the molecular and cellular levels. One of the recent modeling studies we've
done is on priming – subconscious learning that affects your behavior. We
want to know how the information that is constantly flowing into our brain
is integrated into our current knowledge base, and how is that done in such
a way that allows us to recall relevant details and make timely decisions.
It's a complicated problem, but we are making progress by reconstructing
the biophysical and biochemical changes that occur within neurons in the
first seconds of a new memory.
There have been some fascinating technological breakthroughs that have come out of your lab. Can you share some recent examples?
SoftMax, a company that grew out of an independent component analysis
(ICA) algorithm that we developed 10 years ago, worked on noise cancellation
in cell phone headsets. It involved two microphones, using the two
signals to cancel out the background noise that makes it difficult for the
person who is trying to listen to you. It was so successful that Qualcomm
bought the whole company for the technology and the engineering talent,
which included one of my former postdocs. My lab went on to apply ICA
to electroencephalogram (EEG) recordings from the scalp and functional
Magnetic Resonance Imaging (fMRI), which is now used routinely in
hundreds of labs around the world.
Now a new company, NeuroVigil, is analyzing EEG to score automatically the various stages of sleep. One of the bottlenecks in the sleep labs around the country is that it takes a human expert about four hours to analyze eight hours of sleep EEG. But with the technology we've developed we can now do that in about 10 seconds with higher accuracy. It's like a microscope for looking into sleep. Ultimately, we think it will have a major impact on the way sleep is understood, both scientifically and from a medical perspective, to help patients who have sleep disorders and to diagnose other diseases.
In your career, what has been the discovery that has fascinated you most about the brain?
About 10 years ago, two very talented postdocs in my lab made a
spectacular breakthrough that started with a simple model for honeybee
foraging. Bees are risk aversive and they're very good at learning. That
initial paper came out in Nature and led to a breakthrough in understanding
human reward learning. We went on to model the dopamine neurons
in the mammalian brain and showed that the very same algorithm that
we used for the bee also works for the human, which is the basis for a
whole new field called Neuroeconomics. What was fascinating was how a
relatively small insight into something as obscure as honeybee learning
could lead to a new understanding of the brain and behavior. And it
applies not just to reward learning, but also to addiction because every
addictive drug works through the dopamine pathway that we identified
as being important for predicting future reward. Now that we know how
this works, it may lead to much better insight into what motivates us
and the limits of free will.
You and Francis Crick were friends. Please share some of your memories of him?
My wife and I were close to the Cricks and often did
things socially. I remember when Francis was sick
with cancer, we took him and Odile out to see the
movie "Crouching Tiger, Hidden Dragon." It was a
fun movie. I don't think it was Francis' cup of tea,
but I'm sure he appreciated the opportunity to take
some time off from worrying about his health.
He would also come to my lab's tea every day.
He loved discussing science. It was great because
we benefited from his insights and wisdom. I missed
him when his health declined and he was only able
to come to the Salk for an hour or two in the morning.
He was very generous. I think what was remarkable
about him was that if you had something interesting
to say it didn't matter who you were. What mattered
was the science.
What's the best piece of advice he ever gave you?
He once told me that I had become too fond of
my computer models, and that I had lost track of
the fact that the purpose of a model is to make
predictions and to set up a killer experiment that
no one else would ever have thought of doing. I
think he was right in that creating a model often
does become an end in and of itself. He had a
healthy attitude. He saw the value of many different
approaches. One of the main reasons why I came to
Salk was that he appreciated what I was doing
long before a lot of others did.
You organize several workshops each year, including the annual Neural Information Processing Systems Conference. You've authored several books in your field and you're the founding editor of the journal Neural Computation. What drives you to stay so busy?
It's not so much what drives me — it is what comes
naturally. What drives you to eat and sleep? It is
part of my natural rhythm. I enjoy getting people
together, so I organize workshops and meetings. I
have a great lab and it's a great source of pleasure
and pride that so many of my former students are
doing well. I was a postdoc with Steve Kuffler
at Harvard, who was famous for developing new
preparations and encouraging students to carry on
with them and build their own careers. If a single
lab produces a single offspring, that's replacement.
My lab probably has produced about 25 first-rate
researchers who now have their own labs, which is
way outside the normal curve. So my impact is
probably going to be far greater through the students
I've trained than by the papers I've published.