We recently sat down with co-lead instructors and co-founders, Mark Reimers and Pascal Wallisch, to talk about their biennial course on Neural Data Science. The course was first offered in 2015 to help neuroscientists develop conceptual and practical capabilities for analyzing large datasets such as those resulting from single- and multi-electrode extracellular recordings, local field potentials and electroencephalograms (EEGs), as well as two-photon and wide-field optical imaging. Cutting edge data analysis methods are still very much the focus of the Neural Data Science course. Here, Mark and Pascal go into detail about the course’s importance to neuroscience.

Pascal: We now have a lot of new ways of recording, storing, and processing neural data but we don’t yet have a good understanding of how to analyze the data. The richness of the newly-available data demonstrates how much we still don’t understand about the brain and that one would be well-advised to be more modest and humble about what we actually understand about brain function. A lot of other courses focus on existing theories and models; whereas our course de-emphasizes modeling and theory. We won’t tell you what to think about how the brain works, but rather how to analyze your data. 

Mark: I would agree that this course really is very distinct from other computational neuroscience courses. We’re not focused on trying to simulate the brain or impose models on data. We’re exploring the data in a hypothesis-free context rather than use it to buttress or falsify a particular theory. I think the real imperative for neuroscientists now is to look closely at their data and to let it inform them, rather than impose their ideas on their data. Because the new technologies enable us to see more richly than before, I would argue that it’s appropriate in this stage of neuroscience to be more exploratory. 

Pascal: I think the students are loving the exploration aspect of it. It’s a very hands-on course. Someone came up to me this year and said, “This course instills in me confidence.” He is trained as a medical doctor with no mathematical background and was always afraid of equations. But our hands-on approach showed him that he can absolutely do this. It’s actually not that hard once one gets the hang of it.

Mark: The students are asking questions about their data that we can show them how to address. Many of them feel that some of the longstanding theories and ideas are not really giving them insight or traction on their data. So I think they’re welcoming of this new perspective.

Pascal: Regardless of whether someone takes this course or not, those continuing in neuroscience will, in my opinion, need a solid understanding of statistics and at least one relevant programming language to be successful in the field. 

Mark: I second that. I think it’s a fundamental transformation in all of biology, though it’s been particularly rapid and, perhaps, hard for some people at this point in neuroscience. It used to be that it was important for you to have understanding of the animals, ability to do careful surgery, really good microscope skills, be able to identify neurons, and be familiar with your technical apparatus and its artifacts. All of these are still very important but, increasingly, you also have to think quantitatively and pose questions that can be addressed by data analytic methods.

Pascal: Which may be challenging for someone who has no training in that kind of skillset. Biology used to be a field where you could have a long, productive and respectable career while skirting math and statistics for the most part. But this is no longer the case – at least not in neuroscience. So a course like this is desperately needed. 

For the second iteration of the course this summer, Mark and Pascal made revisions to the curriculum and schedule that maximized what the trainees learned. 

Pascal: We made four big changes to this year’s course that made things run much more smoothly. The first is that the course now runs a full two weeks instead of just ten days, with one week dedicated to data recorded with electrical methods and the other to data recorded with optical methods. The second change is that we cut topics that were not directly related to electrical and optical methods, like information theory and fMRI, as there are other courses that focus more directly on those topics. The third change we made is that we streamlined the way we teach, in the sense that we now work with much fewer data sets. In 2015, each invited lecturer brought his/her own data, and we spent hours and hours learning the structures of all the different data sets. Finally, we increased the number of teaching assistants in the course. 

Mark: Most of our theories are built on a kind of information processing model that increasingly seems unlikely to be true except for a few specific areas, such as the visual system, so we no longer emphasize information theory in the curriculum. In addition, we’ve oriented the course towards the new revolutionary technologies like optical imaging. Our plan for the future is to try and pivot toward emerging transformational technologies as they become important. They all pose dramatically different (and difficult) data analysis issues.

We then switched gears and talked about what a typical day for their trainees look like.

Mark: We typically have intense conversations about neuroscience over breakfast, occasionally about politics as well. Then starting at 9 o’clock, we have three hours of seminars, followed by a lunch break. We bring in one invited lecturer to give two talks each morning. The lab sessions then run from 1 to 4:30 and when we say lab, we don’t mean pipetting and test tubes.

Pascal: We mean MATLAB. A key feature of the afternoon lab sessions is that the students work in pairs, and the partners are assigned by us based on their experience with MATLAB. We purposely assign two students per computer so they have someone to talk to while they’re doing analyses and grappling with the data. It prevents the students from feeling isolated and it’s been working really well.

Mark: In the late afternoon, the students have free time when they can go to the tennis court or beach, or take a walk or run around the estate. Dinner is at 6 and, of course, we have even more intense conversations about neuroscience, politics, teaching, and universities. We resume again at 7:30 with evening sessions that are more of a diversion. For example, we’ll have a presentation on how to write a good scientific paper, a discussion about contentious issues in data analysis, or a presentation by someone from Mathworks on new capabilities in their program. The students typically stay up late to do more coding or work on what they didn’t finish during the day. 

Acceptance into the Neural Data Analysis course is competitive. We asked Mark and Pascal what they look for in potential course trainees and their application materials. 

Mark: We’re looking for someone who’s wrestling with fundamental questions about the brain, who’s exploring and using some of the cutting-edge technologies, and who is quantitatively-minded but doesn’t have the tools to analyze the data they’re getting in any detail. We would like to help our students formulate rigorous statistical and mathematical approaches for answering questions that may be lying underneath their experiments, that they want to address but don’t know how to formulate. 

Pascal: We look for three things: 1. Scientists who will benefit from the course. We’ve rejected some applicants because we felt they already knew too much about the material. 2. People who are working on data from electrical or optical methods. Some people who applied for the course were qualified, but were interested in clinical neurology, neurosurgery, or neuro-imaging, and there are other courses for developing those skills. 3. And, of course, we look for overall curiosity and a demonstrated record of achievement. 

Mark: For those whose applications were not accepted this year, I would encourage them to try again. The competition this year was stiff and there were many talented students we couldn’t accept. 

Pascal: We only have 20 spots and we’re already at maximum capacity, so we had to make some really tough choices, heartbreaking in some cases. 

We ended by asking what their favorite moments have been throughout the two iterations of the course.

Mark: During the 2015 course, I fondly remember singing math songs. 

Pascal: I remember that! That was definitely a community building moment. 

Mark: Several of our perhaps best-remembered moments were those where students see that someone is taking seriously a fundamental confusion or uncertainty of theirs, something they had but never articulated very clearly, or had never seen articulated explicitly. Before the course, they’d just had to cope, sort of make do, make stuff up, or do something to produce what’s expected. And now they see that yes, something is really problematic: that it’s not clear how to go from, let’s say, an electrical trace to a discrete set of times where a particular set of neurons has fired. We really need to make processes of the brain visible to researchers so they can come up with better ideas of how the brain works.

Make sure to read the rest of our A Word From... series.