Statistical inference for multi-timescale adaptive threshold and neural mass models via Approximate Bayesian Computation
Sprache des Vortragstitels:
Englisch
Original Tagungtitel:
Statistical inference for multi-timescale adaptive threshold and neural mass models via Approximate Bayesian Computation
Sprache des Tagungstitel:
Englisch
Original Kurzfassung:
In neuroscience it is of primary interest to decode or reconstruct the unobserved signal based on some partially observed information. This corresponds to estimate model parameters of an unknown coordinate based on discrete observations of one or more other coordinates. Quite often, due to the complexity of the models, the underlying likelihood is unknown or intractable, requiring the investigation of new ad-hoc mathematical and statistical techniques to handle it. Here I focus on Approximate Bayesian Computation (ABC) method, and I illustrate it in the framework of stochastic modelling of single neuron and neural network dynamics.
First, I consider the multi-timescale adaptive threshold model, a bivariate stochastic process that can be derived from the detailed Hodgkin-Huxley model, can accurately predict spikes and incorporate the effects of slow K+ currents, usually mediating adaptation. When estimating the underlying parameters, four difficulties arise: none of the two model components is directly observed; the considered process is not of hidden Markov model type; the underlying likelihood is unknown; consecutive spikes are neither independent nor identically distributed. I show how to estimate the threshold parameters only from extra-cellular recordings, i.e. when the spike times are observed.
Second, I consider a stochastic version of the Jensen and Rit Neural mass model, a six-dimensional stochastic process describing the average properties of the electrical activity of a whole population of neurons that has been shown to reproduce EEG data. We are interested in estimating two physiological relevant parameters from partial observations of the model. Experimentally, the process is only partially observed through the EEG-related process, making the statistical inference more challenging. We introduce a Structure-Preserving ABC model taking advantage of the dynamical and structural model properties and validate it on both simulated and real EEG data.