Data assimilation of membrane dynamics and channel kinetics with a neuromorphic integrated circuit

Techniques of data assimilation (DA) for parameter identification and forecasting in complex dynamical systems offer promising tools for the analysis of neural data and inference of neural function. Here we present experiments using DA to characterize the dynamics of a neuromorphic very large-scale integrated (VLSI) circuit emulating membrane dynamics and channel kinetics in a network of 4 generalized Hodgkin-Huxley neurons coupled through 12 conductance-based chemical synapses. The analog VLSI chip, NeuroDyn, features 384 digitally programmable parameters specifying for all neurons and synapses reversal potentials, conductances, and spline-regressed voltage dependence profile of opening and closing rates of the gating variables. In a set of experiments, we conducted DA on the membrane potentials of neurons recorded on the chip under current injection according to the model structure upon which the chip was designed and the known current input sequence, to arrive at the programmed parameters save for model errors due to analog imperfections in the chip fabrication. In a related set of experiments we extended the DA procedure to map songbird neural dynamics onto the chip by identifying and programming parameters extracted using DA from intracellular neural recordings. Application of the chip to neurological data may help to understand the effects of neuromodulators or neurodegenerative diseases on ion channel kinetics, and may further provide insights into the relationship between molecular properties of neurons and the emergence of different spike patterns or different brain behaviors.