Deep learning frameworks for reliable spike sorting and analysis of in vitro neuronal electrophysiology

Neuronal networks of human-derived neurons represent an emerging opportunity to accelerate research on brain function and neurological disease states. Electrophysiological signals recorded from neuronal networks provide distinctive information on the nature of the network and can be employed to unravel mechanisms of relevant neurodegenerative diseases. The complex architecture of excitatory and inhibitory connections thereby governs the selectivity of neuronal responses that can be observed from the network. Hence, a key step in uncovering the functional connectivity within these networks resides in the accurate analysis of how a single neuron can modulate the circuit function and behavior.

High-density microelectrode array recording devices with inter-electrode pitches smaller than 20 mm have greatly improved our understanding of these neural circuits and their activity at the single cell level. In contrast to intracellular single neuron recordings, the individual electrodes are detecting the dense overlap in electrical fields between nearby neurons. To extract the action potential firing times of individual neurons, a computational process known as spike sorting should be applied to assign detected spikes to each individual neuron. Despite great progress in sorting algorithms that cluster spikes with similar waveform shapes, spikes from a single neuron are often assigned to multiple putative clusters.

In this PhD project you will explore algorithmic approaches for real time spike detection and sorting informed by action potential propagation trajectory analysis. Deep learning methods will be used to facilitate the reconstruction process in noisy recording data or to account for time varying biological events. Adaptive selection of microelectrode array recording electrodes will be combined with closed loop experimental feedback to maximize the rate at which electrical profiles can be derived from single neuron and neuronal network behavior. Novel experimental paradigms will be explored to infer network connectivity patterns that are based on external stimulation with different modalities (e.g. mild voltage stimulation or optogenics).

For this interdisciplinary project, you should have a background in computer science, neuroscience, engineering, biomedical engineering, physics, mathematics, machine learning, or related fields. Experience in electrophysiology is considered an asset. Programming experience, primarily Python and Matlab, are required. Prior experience with image processing and computer vision are considered a strong asset. You are a team player and great collaborator. You will be supervised and supported by a team of data/computer scientists, engineers and biologists at imec.

References:

Pachitariu et al 2024 Nat. Methods

Van der Molen et al 2024 Plos One