The goal of our research is to understand how different cell types contribute to computations in the neural circuits of the early visual system how these computations break down in disease. To this end, we use machine learning and statistical modeling to integrate large and complex datasets to identify individual circuit elements, understand how they are connected and link their anatomical and biophysical properties to the computations they perform. Our primary focus is the retina, but we also explore other systems. Finally, we are also interested in exploring machine learning tools for individualized diagnostic of eye diseases.
Data analysis techniques for complex data
We develope techniques to interpret and analyze large, complex dataset in systems neuroscience, including two-photon calcium recordings and single cell transcriptomics. These tools include neuroinformatics resources for data-postprocessing, new methods for spike inference and methods to link neural computations to the expression of specific genes in single cells.
Cell types and connectivity in the retina
We work on functionally classifying the different types of bipolar and ganglion cells in the retina, together with the lab of Thomas Euler. We build the neccessary machine learning algorithms that allow making the best use of the different types of data. We have used successfully used this approach to characterize bipolar (Current Biology, 2013) and ganglion cell types (Nature, 2016).Also, we are interested in the connectivity between the different cell types and cell classes in the retina. For example, we map the connections between photoreceptors and bipolar cells using electron microscopy data. Also, we study how different bipolar cells contribute to the feature selectivity of ganglion cells, funded by the DFG.
Cell types and connectivity in visual cortex
I am working on classifying the building blocks of cortical microcircuits in mice in collaboration with the lab of Andreas Tolias. To this end, we analyze large-scale morphological data acquired by multipatching as well as single-cell transcriptomics.
Machine learning for diagnosing eye diseases
In collaboration with the Zeiss Vision Science lab we develop machine learning algorithms to diagnose eye diseases such as diabetic retinopathy based on clinical measurements such as fundus images. We are particularly interested in providing estimates of the decision uncertainty and exploiting these estimates to improve medical decision making.
Open-source software for statistical computing
We wrote an open-source toolbox for the statistical analysis of angular variables, containing descriptive and inferential statistical methods especially suited for analyzing angular variables (Journal of Statistical Software, 2009). The toolbox is heavily used, demonstrated by more than 200 downloads per month and more than 750 citations to the paper. We also contributed to the development of DataJoint, a tool for managing complex biological data.
- 2017-2020: Landesstifung Baden-Württemberg, Neurorobotik Initiative, “RetNetControl” with G. Zeck
- 2017-2021: DFG CRC 1233 “Robust Vision”, Project 13 with L. Busse and T. Euler
- 2016-2021: BMBF Bernstein Award, “Cell type specific computations”, (FKZ 01GQ1601)
- 2015-2017: DFG Research Grant, “How response diversity of retinal ganglion cells arises from the synaptic interactions in the inner retina” (BE 5601/1-1)
Information about previous research projects can be found here.