From neural genotypes to neural computation: bridging genes, firing patterns and functional excitability classes

Hosted by: Paul Pfeiffer, Louisane Lemaire

Neurons come in many different flavors – neuronal cell types differ in gene expression and morphology, exhibit specific connectivity patterns and show a wide variety of firing behavior. Focussing on the diversity of spiking behavior, here we present recent results on how cortical cell types are linked to specific computations based on a principled mathematical classification scheme and on how genes determine the various firing patterns by combining modern patch-sequencing and machine learning techniques.

External speaker

Yves Bernaerts

Position: Champalimaud Foundation

Title: From code to current to circuitry: bridging multiple levels of modeling in computational neuroscience.

Computational models in neuroscience typically capture different scales and levels of brain function, yet rarely attempt to jointly model them. Here, I present a recently developed multiscale approach to a neuron’s gene expression pattern and firing behavior based on a biophysical-statistical hybrid model. The biophysical part comes from a Hodgkin-Huxley-based model that has been fitted to a variety of mouse motor cortical cell types with simulation-based inference. The statistical part comes from sparse reduced-rank regression, a biologically interpretable and linear technique that we can use to predict the fitted model parameters from a low-dimensional transcriptomic latent representation computed on the basis of few selected ion channel genes. Further, as an attempt to bridge single-cell models to models of neuronal populations, in which neuronal identity is often neglected, I hope to spark a discussion on a new design for latent variable models used to predict population firing rate or animal behavior over time. Here, instead of modeling the loading matrix as a constant parameter to be estimated, it would be modeled as a random variable sampled from a distribution that embraces the rich repertoire of neuronal identity discussed in this symposium.

Lab website: https://fchampalimaud.org/research/groups/memming-park

 

Local Speaker

Paul Pfeiffer

Position: Postdoc with AG Schreiber, HU Berlin

Title: Investigating the relationship between biological and computational neuron types in open databases of neural recordings

Neurophysiologists have early on observed that there are different classes of spiking patterns: the classical example are the distinct fI curves of class 1 neurons that start spiking slowly versus class 2 neurons that start spiking at a minimum non-zero frequency. Mathematically, these experimental observations correspond to the fact that neuronal spiking originates from a small set of excitability classes, each with specific computational properties (e.g. bistability, resonance, etc.). Which neuron adapts which excitability type is hence expected to crucially shape network dynamics and computation, but to date, the “biological” distribution of mathematical excitability classes is unknown. To address this question empirically, I show how to link cortical cell types to excitability classes using the Allen Brain Atlas cell database. Finally, towards a more normative approach, I discuss ideas to use recently developed optimization techniques for spiking neural networks to learn optimal distributions of neural excitability types.

Website: https://orcid.org/0000-0001-5324-5886

Lab website: https://www.neuron-science.de/home

 

 

-----------------------------

 

This event is part of the PostDoc Network Speaker Series.

Guests are welcome!