Through careful measurement and consideration of the brain signals we record, we can build massive databases of brain activity.
We can link these with other open neuroscience data sources to mine for links between brain activity, connectivity, gene expression, cell type, and function in new ways.
We can leverage data mining to generate new hypotheses *for us*.
Our relevant papers:
oscillation parameterization:
https://www.biorxiv.org/content/early/2018/04/11/299859
waveform shape analysis:
https://www.biorxiv.org/content/early/2018/04/16/302000
hypothesis generation:
http://voyteklab.com/wp-content/uploads/Voytek-JNeurosciMethods2012.pdf
open data ecosystems:
http://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1005037
1. BradleyVoytek, Ph.D.
UC San Diego
Department of Cognitive Science
Neurosciences Graduate Program
Halıcıoğlu Data Science Institute
bvoytek@ucsd.edu
@bradleyvoytek
2.
3. Support
• Alfred P. Sloan Research Fellowship in Neuroscience
• Whitehall Foundation
• NIMH R01 MH095984
• NSF BCS COGNEURO 1736028 (Neuroscience)
• NSF DGE NRT 1735234 (Data Science)
• Kavli Innovative Research Grant
• NIHT32 UCSD Institute for Neural Computation Cognitive NeuroscienceTraining Grant
• NIHT32 UCSD Genetics Training Program
13. Oscillations are bumps
“In a power spectrum, brain oscillations
appear as bumps on top of this 1/f
slope…” - He B, Trends Cogn Sci 2014
“[when] particular oscillation frequencies
become dominant… a peak (bump)
appears in the respective frequency band.” -
Buzsáki el al., Neuron 2013
67. fit aperiodic (1/f) in PSD
frequency (Hz)
power
10 20 30 40
remove aperiodic signal
A
B
iteratively fit and remove
Gaussians
C
(halt fitting at noise floor)
multi-Gaussian fit using
iteration parametersD
remove Gaussians from
original PSDE
re-fit aperiodicF
combine
assess goodness of fit
G
H
peak
peak
peak
1/f fit
final fit
Haller, Donoghue, Peterson, et al., bioRxiv
68. Parameterizing neural power spectra
Source: Haller, Donoghue, Peterson, et al., bioRxiv Code: https://github.com/voytekresearch/fooof
95. We can look for connections
between all of these—cell types,
gene expression, function,
oscillations, connectivity—to see
how they all interrelate.
96. And we can find gaps—missing
connections where topics should be
related but aren’t.
97. We can mine through these
connections to find potential
missing links.
98. That is, we can use the data
themselves to generate new
hypotheses for us.
99. BradleyVoytek, Ph.D.
UC San Diego
Department of Cognitive Science
Neurosciences Graduate Program
Halıcıoğlu Data Science Institute
bvoytek@ucsd.edu
@bradleyvoytek