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
P103: Constructing model surrogates of populations of dopamine neuron and medium spiny neuron models for understanding phenotypic differences

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

Feedback form is now closed.
Tim Rumbell, Sushmita Allam, Tuan Hoang-Trong, Jaimit Parikh, Viatcheslav Gurev, James Kozloski

Neurons of a specific type have intrinsic variety in their electrophysiological properties. Intracellular parameters, such as ion channel conductances and kinetics, also have high variability within a neuron type, yet reliable functions emerge from a wide variety of parameter combinations. Recordings of electrophysiological properties from populations of neurons under different experimental conditions or perturbations produce sub-groups that form “electrophysiological phenotypes”. For example, different properties may derive from wild-type vs. disease model animals or may change across multiple age groups [1].

Populations of neuron models can represent a neuron type by varying parameter sets, each able to produce the outputs of a recording, and all spanning the ranges of recorded features. We previously generated model populations using evolutionary search with an error function that combines soft-thresholding with a crowdedness penalty in feature space, allowing coverage of the empirical range of features with models. The technique was used to generate a population of dopamine neuron (DA) models, which captured the majority of empirical features, generalized to perturbations, and revealed sets of coefficients predicted to reliably modulate activity [2]. We also used this technique to construct striatal medium spiny neuron (MSN) model populations, which recapitulated the effects of extracellular potassium changes [3] and captured differences in electrophysiological phenotype between MSNs from wild- type mice and from the Q175 model of Huntington’s disease. Our approach becomes prohibitively computationally expensive, however, when we seek to produce multiple populations that represent many phenotypes from across a spectrum. For example, to recreate the non-linear developmental trajectory observed across postnatal development of DAs [1] we would need to perform multiple optimizations.

Here we demonstrate the construction of model surrogates that map model parameters to features spanning the range of multiple electrophysiological phenotypes. We sampled from parameter space and simulated models to create a surrogate training set. Using our evolutionary search as prior knowledge of our parameter space enabled a dense sampling in regions of the high- dimensional model parameter space that were likely to produce valid features. We trained a deep neural network with our datasets, producing a surrogate for our model that maps parameter set distributions to output feature distributions. This can be used in place of the neuron model for model sampling, allowing rapid construction of populations of models that match different distributions of features from across multiple phenotypes. We demonstrate this approach using DA developmental age groups and MSN disease progression states as targets, facilitating a mechanistic understanding of parameter modulations that generate differences in phenotypes.

References

1. Dufour, M.A., Woodhouse, A., Amendola, J. et al. (2014) Non-linear developmental trajectory of electrical phenotype in rat substantia nigra pars compacta dopaminergic neurons. eLife. 3:e04059.
2. Rumbell, T.H. and Kozloski, J. (2019) Dimensions of control for subthreshold oscillations and spontaneous firing in dopamine neurons. PLoS Comput. Biol., 15(9): e1007375.
3. Octeau, J.C., Gangwani, M.R., Allam, S.L., et al. (2019) Transient, consequential increases in extracellular potassium ions accompany channelrhodopsin2 excitation. Cell Reports. 27: 2249-2261.

Paper links
https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1007375
https://www.biorxiv.org/content/10.1101/2020.06.01.128033v1



Google Meet Link
https://meet.google.com/qom-htnd-iaq

Speakers
TR

Tim Rumbell

Healthcare and Life Sciences, IBM Research



Sunday July 19, 2020 9:00pm - 10:00pm CEST
Slot 20