Description:
Cell neurobiologists know a good deal about individual neurons, the signaling pathways between pairs of neurons, and the dynamics of small networks of neurons. Cognitive neuroscientists, on the other hand, can tell you a lot about the behavior of organisms and the parts of their nervous systems that appear to be responsible for governing that behavior. There is, however, relatively little known about how microscale molecular processes implement the computations that underlie cognition and ultimately cause observable macroscale behavior.
A major goal of computational neuroscience is to produce predictive mesoscale theories of biological computation that bridge the gap between the cells and behaviors of complex organisms thereby explaining how the former give rise to the latter. Until recently there was little hope of formulating testable theories of this sort. However, with new technologies for recording the activity of thousands, even millions of neurons simultaneously, it is now feasible to observe neural activity at a scale and resolution that opens the possibility of inferring such theories directly from data.
In this year's class, we continue our investigation of functional connectomics, the study of how the structure of neural circuits and the activity of their constituent neurons perform the functions necessary for cognition and control [7, 8, 10, 5, 4]. In the last year alone, dozens of papers have been published describing new methods for inferring function from recorded structure and activity. A number of the authors of those papers will participate in class to describe their methods and share their models and data. Students taking the class for credit can apply these models or those of their own devising to learn models from neural recordings provided by participating labs.
To whet your appetite, we consider two rather ambitious and technically demanding approaches to big data in neuroscience. The first approach employs technologies from synthetic biology to construct a physical model of a neuronal tissue sample using experimental protocols from neurobiology and biochemistry developed in Ed Boyden's lab [1] at MIT and Mark Ellisman’s lab [9, 3] at UC San Diego. Here is a prospectus of the sort one might pitch to a venture-capital firm to fund a startup developing such technology. In class, we'll explore if this is even possible, much less practical in the near future.
The second approach applies techniques from computer science and machine learning in attempting to build a mesoscale model bridging the gap between molecules and behavior. The pitch in this case features two papers [6, 2] describing the work of two groups of scientists. One paper describes a model of worm movement involving a few hundred neurons. The other paper presents a model of rat vibrissal (whisker) touch involving tens of thousands of (simulated) neurons in the somatosensory cortex.
The prerequisites are basic high-school biology, good math skills, and familiarity with machine learning. Some background in computer vision and signal processing will be important for projects in structural connectomics. Familiarity with modern artificial neural network technologies is a plus for projects in functional connectomics.
Location and Time:
MW, 4:30-5:50pm, Building 100, Room 101K
Staff:
Instructor: Thomas Dean
Email: tld [at] google [dot] com
Office hours: by appointment
Course Assistant: TBD
Email: TBD
Textbooks:
There are no required textbooks for this course but
you are expected to do a lot of reading on your own
and these three texts are good to have around for
reference. I’ve yet to meet anyone who has
read all three cover to cover but over the years
I’ve probably read most of the chapters in one
edition or the other and often found them relevant.
-
- Neuroscience: Exploring the Brain (Third Edition), Bear, Connors and Paradiso.
-
- The Cognitive Neurosciences (Third Edition), Gazzaniga.
-
- Principles of Neural Science (Fourth Edition), Kandel, Schwartz and Jessell.
Grading:
- Class participation including presentation (30%)
- Project proposal due around midterm (20%)
- Project report due around finals week (50%)
References: