Data-Driven Network Neuroscience Sarah Feldt Muldoon Mathematics, CDSE Program, Neuroscience Program DAAD, May 18, 2016
What is Network Neuroscience?! Application of network theoretical techniques to neuroscience data in order to quantify and understand brain structure and function.
What is a Network? A network is a collection of nodes (vertices) connected by edges (links). 2 1 4 6 5 nodes edges Adjacency Matrix 3 7 8 9 12 10 11
Real-world Networks: Systems to Graphs Dolphin Social Network Nodes: Bottlenose dolphins Edges: Social interactions Map of Science Nodes: Journals Edges: Click stream data Lusseau (2007) Evol Ecol Food Flavor Network Nodes: Ingredients Edges: Shared flavor compounds Bollen et al. (2009) PLoS ONE Ahn et al. (2011) Scientific Reports
Fun Brain Facts The human brain has ~10 11 neurons Each neurons has ~ 10 4 synapses Your brain has fired ~ 10 13 action potentials in the last minute
Types of Neuroscience Data Recent explosion in experimental techniques Micro-scale - + Macro-scale Structural data Time series data (functional data) Simultaneous recordings across modalities Multi-electrode Array Spatial resolution Temporal resolution ECoG Data across multiple scales Ca imaging + - fmri How do we integrate across scales, modalities, and temporal evolution?
Networks in Neuroscience Choice of scale determines how networks are built Micro-scale Macro-scale Neurons Brain Regions
Defining Nodes: Choice of Scale Micro-scale: Neurons Meso-scale: ECoG sensors, probes Excitatory or Inhibitory Macro-scale: Brain Regions Anatomically defined Functionally defined
Defining Edges: Anatomical Networks Structural networks: reflect the underlying anatomy Micro-scale: synapses (chemical transmission) or gap junctions (electrical) Meso/Macro-scale: white matter tracts between brain regions Pereda, (2014) Nat. Rev. Neuro
Additional Information: Node Dynamics We can take advantage of the fact that we observe node dynamics (underlying structure is often more difficult to determine) Discrete dynamics Neuron 1: Neuron 2: Neuron 3: Time Continuous dynamics
Defining Edges: Functional Networks Derived from statistical relationships between node dynamics Discrete spike train data Neuron 1: Neuron 2: Neuron 3: Time Continuous data: electrodes, fmri, MEG, etc. Statistical Relationships (correlations) Functional Connectivity
Defining Edges: Other Considerations Choice of frequency band Bandpass filter and operate in time domain Choose measures that operate in frequency domain: Coherence, wavelets Weighted vs Binary Networks Telesford et al. (2013) Front Neurosci Thresholding can potentially throw away important information about network structure Comparing graphs: van Wijk et al. (2010) PLoS ONE Weak connections in schizophrenia: Bassett et al. (2012) NeuroImage
Why Study Brain Networks? Compare structure between groups Healthy vs. pathological: Identify biomarkers Understand individual differences How does structure give rise to function and performance? Study network evolution How does structure change over time (learning, disease progression, etc) Understanding network organization can provide insight into brain function (dysfunction)
Data-driven Network Neuroscience
From Data to Networks Micro-scale data Calcium imaging Muldoon et al. (2015) Brain Goldberg Lab, U of Penn
From Data to Networks Meso-scale data ECoG Window 1 Window 2 Window 3 Window 4 Nodes (electrodes) 10 20 30 40 50 60 70 80 1 1 2 3 4 5 6 7 Time Window 4 3 2 Community 5 RH54-Ref 6 SD5-Ref Lesser and Webber, Johns Hopkins
From Data to Networks Macro-scale: MRI Data fmri data (task data) Network Structure A Network Structure Nodes: 83 brain regions Dynamics + Edges: white matter tracts Functional Connectivity Jean Vettel, ARL Scott Grafton, UCSB Matt Cieslak, UCSB, Clint Greene, UCSB, John Medaglia, Penn 0 0.5 1
From Data to Networks New Data: Tractography in rodents Poulsen and Schweser, UB: Rat Data
Quantifying Brain Network Structure Many of the techniques and measures used to study brain networks were originally developed for social networks However, there are some important differences between the properties of the data between the two systems Often weighted networks Not necessarily sparse networks Nodes are dynamic Now many methods have been adapted to reflect the properties of neural data, but this is an active area of research that continues to be developed
Necessary skills Data analysis: extracting networks from times series or imaging data Signal processing Image analysis Computational abilities Metric development: novel methods for quantification of network structure tailored to properties of neuroscience data Applied math (networks) Topology Computational abilities Need for researchers with highly interdisciplinary skill sets
Ph.D. Program in Computational & Data Enabled Science (CDS) This is a program that properly addresses the need to cover the two important areas of data and computation in an integrated manner UB is ahead of the efforts that this reviewer has seen to date. 140,000-160, 000 new jobs in CDSE related areas New PhD program MS in discipline required Fellowships cdse.buffalo.edu
CDSE Program Talk tomorrow by Abani Patra 3:05 PM
Thanks to Muldoon Group Michael Vaiana Henry Baidoo-Williams Minwei Ye Kanika Bansal Many collaborators Dani Bassett Lab, U of Penn. Jean Vettel, US Army Research Laboratory Scott Grafton, UCSB Matt Cieslak, USCB Clint Greene, UCSB John Medaglia, Penn Ron Lesser, Johns Hopkins Bob Webber, Johns Hopkins Ethan Goldberg, Penn/CHOP David Poulson, UB Ferdinand Schweser, UB Kostas Slavakis, UB David Wack, UB Mark Bower, Mayo Clinic Funding:
Questions?