PROBLEMA BIOLOGICO

1
Inferring Neural Connectivity and the Underlying
Network Dynamics fromSpike Train Recordings
Valeri A. MAKAROV(1), Oscar DE FEO(2), and Fivos
PANETSOS(1) 1Neuroscience Laboratory, Department
of Applied Mathematics, Escuela de Optica,
Universidad Complutense de Madrid,
Spain. 2Laboratory of Nonlinear Systems,
Swiss Federal Institute of Technology, Lausanne,
Switzerland. Email vmakarov_at_opt.ucm.es
2
Biophysical problem and Experimental design
3
Aim reconstruct circuitry
Exper. Results Spike trains (multivariate point
process) of a number of neurons No information
on network structure, intrinsic dynamics,
synaptic connections etc.!
BUT We want to know all these!!!
4

• Known statistical methods Crosscorrelation,
  Partial Spectral Coherence, Partial Covariance
  Density, Granger causality based methods
• Common problems of statistical methods
• In time domain problems with scaling
  (difficulties with
• rhythmic neurons)
• In frequency domain no
  distinction between inhibitory
• and excitatory
• To get statistic long-lasting
  stationary spike trains

5
Our novel deterministic method
6
Basic idea of the method (illustration for two
neurons)
PART 1 Identification
7
Operating scheme of the identification algorithm
Minimize the difference between model prediction
and experimental spike trains. The cost function
G is inversion of
8

• PART 2 Modeling
• Ones Part 1 is done, we have a DS with all known
  parameters. Thus, we can
• investigate it independently on the original data
  (bifurcations, stability, etc).
• Assess any statistical properties, e.g.
  cross-correlation, ISIH.
• Use it as a basis for constructing conductance
  based (e.g. HH) models.

DS of the network
9
Testing of the method and examples of its
application
10
Example a DCN-like (artificial) network
11
Example recordings from the DCN
Identification
Modeling
12
Conclusions

• We have proposed a novel deterministic method
  capable
• Identifies correctly synaptic coupling
  including
• Direction
• Polarity (inhibitory vs excitatory)
• Mutual couplings
• Indirect couplings
• Works with relatively short spike trains (avoid
  problem of
• stationarity)
• Robust (checked with noisy and regular spike
  trains)
• Allows natural modeling of the neural activity
  (new)
• Allows the use of a priori knowledge (like fast
  neuron is
• inhibitory)
• For success, the method has to be used in the
  context. Anatomical, electro-physiological, and
  experimental design should be taking into
  account.

13
(No Transcript)