Berens Lab

Neural Data Science for Vision Research

Home » Labs » Berens Lab » Research topics

Research Subjects

Data science – turning data into knowledge

We use techniques from machine learning, high-dimensional statistics, neural coding, visualization and data management to make use of large and complex datasets in visual neuroscience, with a particular focus on the retina. Specifically, we aim to link the computation performed by individual cell types in the visual system to their underlying biophysics, morphology and connectivity. Our model systems are the circuits of the early visual system with a focus on the retina primary visual cortex of mice and corresponding structures in the zebrafish visual system. We collaborate closely with our experimental partners to answer fundamental questions about how visual information is processed in these structures.

Cell-type identification based on functional, morphological or transcriptomic data

We are working on methods to identify the building blocks of the visual system, the cell types of the retina and visual cortex, based on large scale functional two-photon imaging experiments, high-resolution morphological reconstructions and single-cell transcriptomes. The complexity and variety of data sources requires specialized statistical models for reliable cell type inference. In particular, we are interested in how one can efficiently combine (partially incomplete) data from different sources.

Functional diversity of OFF retinal ganglion cells in the mouse retina. We used a Mixture of Gaussian clustering approach to identify functional RGC types based on large scale two-photon calcium recordings (reproduced from Baden et al., Nature, 2016).
Three dimensional rendering of type 4 bipolar cells (red) and cone pedicles (blue) in the mouse retina based on high-resolution electron microscopy data. We used this dataset to systematically study the connectivity of bipolar cells to photoreceptors.

Anatomy-guided neural response models

We develop functional models of neural activity exploiting available anatomical data to explain the computations implemented by specific cell types. For example, we obtained high-resolution electronmicroscopy-based reconstructions of the outer retina to determine the synaptic connectivity of different bipolar cell types. We use this information to constrain the space of functional response models for bipolar cell activity. We believe that such models will allow integrating anatomical and functional datasets.

Linking transcriptomic and morphological data to computation

We explore the use of detailed biophysical models of retinal bipolar cells for explaining cell type specific computations. In particular, we believe that such models can provide valuable links between information about gene expression levels acquired in single-cell transcriptomics, high-resolution anatomical reconstructions and functional measurements. In addition, we explore the use of sparse regression techniques to infer physiological properties from single cell transcriptomic measurements acquired using patch-seq. 

Reconstruction of five neurons in primary visual cortex of mice recorded with a multi-patching setup and their connections (reproduced from Jiang et al., Science, 2015). We develop machine learning methods to infer cell types and connectivity profiles from such sparse recordings.
Single-cell transcriptome acquired using Patch-seq from cortical layer 1 interneurons (reproduced from Cadwell et al., Nature Biotechnology, 2016).

Effects of degenerative diseases and clinical diagnostics

We are interested in how diseases differentially impact different cell types throughout the visual system. In the retina, we study whether certain cell types throughout the retina are particularly susceptible to degenerative diseases such as retinitis pigmentosa. In addition, we search for markers in signals such electroretinograms predictive of changes on the cellular level. We are also interested in how this work may have consequences for treatment of such diseases and the design of neuroprosthetic devices and how we can use our techniques to more effectively diagnose degenerative eye diseases in humans.

Reproducible science

We believe that the best way to achieve reproducible scientific results is to promote open science, including sharing data and making software for scientific research freely available.