Complex systems neuroscience

What questions one should ask in order to reverse engineer the brain?
How does one do exploratory neuroscience without a preconceived hypothesis (and is this even possible)?
What do you do with the data if you can record from 1000 neurons in a single go?
How do you analyse task free, naturalistic behavior?
What happens if you use modern machine learning methods to try and predict the video recording of an animal behaving using as input the spiking activity from 1000 neurons?

A first approach to answering some and a few more of the above questions using very large scale electrophysiology experiments and naturalistic behavioral tasks.

Transcript

1. Complex systems neuroscience George Dimitriadis
2. Reverse engineering complex systems Observation of information exchange Disassembly Decompilation ?
3. Recording dimensions in systems neuroscience 32 channels Neuronexus commercial 128 channels Neuroseeker project IMEC design 384 channels Neuropixels project IMEC design 4 channels hand made 1344 channels Neuroseeker project IMEC design 1440 sites, 8 mm 100µm
4. Behavioural dimensions in systems neuroscience Behavioural variables = One Hypothesis driven Anne Marie Brady, Stan B. Floresco Behavioural variables = A few Hypothesis driven Ruth A. Wood et al. Behavioural variables = Many Exploratory driven Behavioural variables = Many Both hypothesis and exploratory driven Carsen Stringer et al.
5. Summary • High dimensional recordings: Design concepts and validation • High dimensional behavioural setup • Putting it all together • Speed correlations • Encoding salient features of a world model • Exploratory analysis
6. Summary • High dimensional recordings: Design concepts and validation • High dimensional behavioural setup • Putting it all together • Speed correlations • Encoding salient features of a world model • Exploratory analysis
7. Neuroseeker probe: Design • 1344 ACTIVE electrodes • Scanning amplifier • 10 bit res. • 8mm x 100um x50um • Electrodes separated in 12 regions • 112 measuring • 8 reference electrodes • Each set to either LFP or AP band One region Reference 22 mm
8. Neuroseeker probe: Recordings
9. Neuroseeker probe: Live
10. Neuroseeker probe: Lifespan and reuse
11. Summary • High dimensional recordings: Design concepts and validation • High dimensional behavioural setup • Putting it all together • Speed correlations • Encoding salient features of a world model • Exploratory analysis
12. Behavioural setup: The machine
13. Behavioural setup: The experiment (training) Sound to reward availability Touching light to reward availability
14. Behavioural setup: The experiment (running) Touching light Moving light
15. Summary • High dimensional recordings: Design concepts and validation • High dimensional behavioural setup • Putting it all together • Speed correlations • Encoding salient features of a world model • Exploratory analysis
16. Speed correlations Mutual Information Alexander Kraskov et al. number of points with number of points with where Advantages • Corrects for bias due to sample size • Can be used with continuous data (does not require prior binning)
17. Speed correlations
18. Speed correlations
19. Speed correlations Number of neurons
20. Summary • High dimensional recordings: Design concepts and validation • High dimensional behavioural setup • Putting it all together • Speed correlations • Encoding salient features of a world model • Exploratory analysis
21. Encoding salient features of a world model Events definition
22. Encoding salient features of a world model Trial pokes
23. Encoding salient features of a world model Non trial pokes
24. Encoding salient features of a world model Positions of modulating neurons
25. Encoding salient features of a world model Activity modulating neurons in the hippocampus have place fields spanning the whole arena
26. Encoding salient features of a world model Normalised firing activity around salient events for all neurons
27. Encoding salient features of a world model Normalised firing activity around trial pokes events for all neurons over days
28. Summary • High dimensional recordings: Design concepts and validation • High dimensional behavioural setup • Putting it all together • Speed correlations • Encoding salient features of a world model • Exploratory analysis
29. t-Student nonlinear embedding (t-sne) Laurens Van Der Maaten, Geoffrey Hinton Exploratory analysis
30. Exploratory analysis t-sne of spike rate vectors (top 40 primary components / explained Var = 0.22)
31. Exploratory analysis t-sne of behaviour video (top 100 primary components / explained Var = 0.98)
32. Exploratory analysis t-sne of behaviour video clustered with DBSCAN
33. Exploratory analysis Correspondence of behaviour clusters to the spike t-sne
34. Exploratory analysis Using NNs to predict the video: Architecture
35. Exploratory analysis Using NNs to predict the video: Results Image fed to the networks Target image to predict Image predicted by the Images Only network Image predicted by the Images and Spikes network
36. Acknowledgments ATLAS Neuroengineering Arno Aarts Tobias Holzhammer IMEC Andrei Alexandru Marco Ballini Nick Van Helleputte Chris van Hoof Carolina M. Lopez Silke Musa Bogdan Raducanu Jan Putzeys Shiwei Wang Marleen Welkenhuysen RADBOUD U. Francesco Battaglia Eric Maris Tim Schroeder Paul Tiesinga CNRL Francois David Luc J. Gentet ICNP Richárd Fiáth Domonkos Horváth Gergely Márton Domokos Meszéna Istvan Ulbert IMNS Srinjoy Mitra UPPSALA U. Hercules Neves U. OF PARMA Guy A. Orban IMTEK Frederick Pothof Patrick Ruther U. Of CAL. IRVINE Bruce L. McNaughton ESIN Wolf Singer Spyros Samothrakis Joana Neto Adam Kampff Atabak Dehban Lorenza Calcaterra Joana Nogueira Gonçalo Lopes Andre Marques-Smith
37. Joana Neto Adam Kampff Atabak Dehban Lorenza Calcaterra Joana Nogueira Gonçalo Lopes Acknowledgments Do not obstruct knowledge. Adam R. Kampff 2018 Andre Marques-Smith
38. Acknowledgments Do not obstruct knowledge. Adam R. Kampff 2018 The rule of reason: In order to learn you must desire to learn, and in so desiring not be satisfied with what you already incline to think. Corollary: Do not block the way of inquiry. Charles Sanders Peirce c.1890