Machine Learning talk by Christopher Currin, IST Austria

Human cortical cultures and artificial models: Understanding individuals with epilepsy

27 Feb 2024 from 4 p.m. to 27 Feb 2024 5 p.m. CET (GMT+0100)
CWI, room L120

Speaker: Christopher Currin, NOMIS Fellow from the Institute of Science and Technology Austria, working closely with the groups of Tim Vogels and Gaia Novarino there.

Title: Human cortical cultures and artificial models: Understanding individuals with epilepsy

Abstract: Artificial neural networks simplify complex biological circuits into tractable computational models to distil their essence and test our understanding. It is often said that the simplicity of artificial models undermines their applicability to real brain dynamics. Typical efforts to address this mismatch add complexity to increasingly unwieldy models. We instead use simplified cortical cultures with two cortical neuron types derived from human induced pluripotent stem cells (hiPSCs) to compare model and reality.
Over 6 weeks of development, we simultaneously record thousands of neurons using high-density microelectrode arrays (HD-MEAs). Recording at high-density and large scale allows an in-depth look at dynamics at both the neuron and systems levels.
We build unique profiles of dynamics from a library of metrics measured from the 39 cultures. Our approach of deriving neural networks from hiPSCs also allows us, for the first time, to directly compare neural dynamics of epilepsy patients and close family “control” members. We uncovered surprisingly variable network activity across cultures from families with a common genetic variant. That is, no single measure or metric was sufficient to classify a culture according to the genetic background of its donor, even if differences within a family were evident. We try to address this using two complementary methods. First, a multi-metric approach that clusters individuals using dimensionality reduction of their dynamics within the dataset. Second, by building data-driven models – “digital twins”, or artificial reproductions of each network – that allow rapid and reproducible probing of individuals’ dynamics. Furthermore, the digital twins suggest ex vivo perturbation experiments for further understanding of the individuals’ cultures and models.
Our research showcases a promising personalised medicine approach for the understanding and treatment of people with epilepsy. It does so by using data-driven modelling that starts to bridge an important theoretical-experimental neuroscience gap for advancing our understanding of human neuron dynamics.