Marc Büttner: Classification of Retinal Ganglion Cells based on electrical responses recorded with High-Density Microelectrode Arrays

BCCN Berlin / TU Berlin

 

Abstract

The retina is the visual entry point to the brain. Understanding how this complex machinery processes and encodes visual information has been a prominent research topic in neuroscience for centuries. It has been shown recently that the functional diversity of parallel output channels, the retinal ganglion cells (RGCs), in the mouse retina is much higher than previously thought. In this thesis High-Density Microelectrode Array (HD-MEA) recordings of mouse and marmoset RGCs were clustered using unsupervised machine learning techniques. First, based on a linear-non-linear Poisson model, responses from eight known RGC types were simulated, which served as ground truth data for building a clustering pipeline. It was found that applying Uniform Manifold Approximation and Projection for Dimension Reduction (UMAP) to the features did not only deliver good clustering results but did so in low dimensions allowing for visualization and interpretation of the data. Second, the clustering pipeline was applied to real HD-MEA mouse (n=976) and marmoset (n=809) retinal recordings of responses to a Chirpsweep stimulus as well as to another mouse data set (n=418) of responses to a novel Random Moving Objects (RMO) stimulus. Finally, the clustering results were used to analyze the response properties of different types of RGCs. Several distinguishing features in their response profiles were determined across species. Furthermore differences in functional responses between mouse and marmoset RGCs were analyzed revealing that response mechanisms towards gradient frequency modulations of full-field light stimulation could differ across species. The results of this thesis show that the developed clustering pipeline captures key properties of RGC responses and is well suited for clustering and analyzing RGC types of different species.

 

Additional Information

Master Thesis Defense

 

Organized by

Klaus Obermayer / Lisa Velenosi