Loading…
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 • 9:00pm - 10:00pm
P62: Large Scale Discrimination between Neural Models and Experimental Data

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

Feedback form is now closed.
The poster meeting virtual room:
meet.google.com/toi-hoie-pfa

Russell Jarvis
, Sharon Crook, Richard Gerkin

Scientific insight is well-served by the discovery and optimization of abstract models that can reproduce experimental findings. NeuroML (NeuroML.org), a model description language for neuroscience, facilitates reproducibility and exchange of such models by providing an implementation- agnostic model description in a modular format. NeuronUnit (neuronunit.scidash.org) evaluates model accuracy by subjecting models to experimental data-driven validation tests, a formalization of the scientific method.

A neuron model that perfectly imitated real neuronal electrical behavior in response to any stimulus would not be distinguishable from experiments by any conventional physiological measurement. In order to assess whether existing neuron models approached this standard, we took 972 existing neuron models from NeuroML-DB.org and subjected them to a standard series of electrophysiological stimuli (somatic current injection waveforms). We then extracted analogous 448 stimulus-evoked recordings of real cortical neurons from the Allen Cell Types database. We applied multiple feature extraction algorithms on the physiological responses of both model simulations and experimental recordings in order to characterize physiological behavior with a very high degree of detail spanning hundreds of features.

After applying dimensionality reduction to this very high dimensional feature space, we show that the real (biological neurons) and simulated (model neurons) recordings are easily and fully discriminated by eye or any reasonable classifier. Consequently, not a single model neuron produced physiological responses that could be confused with a biological neuron. Was this a defect of the model design (e.g. key mechanisms unaccounted for) or of model parameterization? The remaining post- optimization disagreement between models and biological neurons may reflect limitations of model design and can be investigated by probing the key features used by classifiers to distinguish these two populations.

Speakers
avatar for Russell Jarvis

Russell Jarvis

Neuroscience, Arizona State University
I am interested in Free and Open Source Toolchains, the application of FOS technology to neuronal data sets. I am especially interested in neuromorphic computing, GPU network models, and information theory.



Sunday July 19, 2020 9:00pm - 10:00pm CEST
Slot 08