CNS*2020 Online has ended
Welcome to the Sched instance for CNS*2020 Online! Please read the instruction document on detailed information on CNS*2020.
Back To Schedule
Sunday, July 19 • 8:00pm - 9:00pm
P144: Closed loop parameter estimation using GPU acceleration with GeNN

Log in to save this to your schedule, view media, leave feedback and see who's attending!

Feedback form is now closed.
Google Meets: https://meet.google.com/eku-dfxf-ozz 

Felix B. Kern, Michael Crossley, Rafael Levi, György Kemenes, Thomas Nowotny

  A common approach to understanding neuronal function is to build accurate and predictive models of the excitable membrane. Models are typically based on voltage clamp data where ion channels of different types are pharmacologically isolated and the stationary state and timescale of (in)activation are estimated based on the transmembrane currents observed in response to a set of constant voltage steps. The basic method can be extended with different stepping protocols or input waveforms and by performing parameter fits on the full time series. Further improvements are achieved with parameter estimation on additional current clamp data, an active field of research. Some examples of employed estimation approaches include adaptive coupling to synchronise the model to data, driving neurons with chaotic input signals, and using distributions of parameter values in a path integral method.

Enabled by our GPU enhanced neural networks (GeNN) framework [1], we here present work that makes a different conceptual advance of performing parameter estimation in an online closed loop approach while the neuron is being recorded. In doing so we can select stimulations that are highly informative for the parameter estimation process at any given time. We can also track time dependent parameters by observing how parameter estimates develop over time.

To demonstrate our new method we use the model system of the B1 motor cell in the buccal ganglion of the pond snail _Lymnaea stagnalis_. Neurons are recorded with two sharp electrodes in current clamp mode. We have built a conductance based initial model from a published set of Hodgkin-Huxley conductances [2], using standard parameter estimation methods and data we obtained with a simple set of current steps. To perform closed loop parameter estimation, we use a genetic algorithm (GA) in which a population of 8192 model neurons with candidate parameter values is simulated on a GPU (NVIDIA Tesla K40c) in parallel and in real time. Models are then compared to the response of the recorded neuron and selected for goodness of fit, as is standard for a GA approach. The novel element of our method is the next step, where we evaluate a pool of candidate stimuli against the model population, selecting the stimulus with the most diverse responses for the next epoch. The selected stimulus is then applied to both the recorded neuron and the population of models and the normal GA procedure continues.

Fig. 1 shows a representative example of online fitting to a neuron. We first fit a set of 52 parameters to fine-tune model kinetics to the cell under stationary conditions. Then, we restricted fitting to non-kinetic parameters (maximum conductances, equilibrium potentials, and capacitance) and continued to run the algorithm described above, while at the same time manipulating sodium levels in the extracellular bath. The online fitting procedure can detect and track the change in sodium concentration as putative changes in sodium conductance and reversal potential.

This work was partially supported by the EU under Grant Agreement 785907 (HBP SGA2).


[1] Yavuz E, Turner JP, Nowotny T. GeNN: a code generation framework for accelerated brain simulations. Scientific Reports 2016, 6, 18854.
[2] Vehovszky A, Szabo H, Elliott CJH. Octopamine increases the excitability of neurons in the snail feeding system by modulation of inward sodium current but not outward potassium currents. BMC Neurosci. 2005, 6, 70.

avatar for Felix B. Kern

Felix B. Kern

Research Fellow, IRCN, UTokyo, Japan

Sunday July 19, 2020 8:00pm - 9:00pm CEST
Slot 03