Welcome!

| philippberens

(c) T. Klink, Bild der WissenschaftMy group is using machine learning and data analytics for large-scale neuroscience and medical data science, working on identifing cell types, circuits and computations in the healthy and diseased visual system and improving clinical diagnostics. We are part of the Institute of Ophthalmic Research at the University of Tübingen and affiliated with the Center for Integrative Neuroscience and the Bernstein Center for Computational Neuroscience in Tübingen. Our official lab website can be found here.

Advertisements

Cones contact rod bipolar cells – replicated

| philippberens

replication pang.pngA bit ago, we published a study in eLife mining a publically available electron microscopy dataset in an attempt to quantify the connectivity in the outer retina. We found that rod bipolar cells (which, as the names suggests, are thought to contact rods primarily) also contact cones, and quite frequently so. This was now confirmed independently by the lab of Sam Wu, in an article published in Journal of Comparative Neurology. They also show that these are likely active synaptic sites, as there is mGluR6 expressed there.

New paper on horizontal cells published

| philippberens

DP85nE6XcAEYpcT

Our new paper in collaboration with the Euler lab has just been published by Current Biology. Camille recorded light-evoked Ca2+ signals in horizontal cell dendrites and showed that they reflect the cone opsin gradient. In contrast to the widespread hypothesis that horizontal cells are globally coupled, chromatic preferences in neighboring dendritic tips vary markedly, indicating that mouse horizontal cells process cone photoreceptor input locally. We contributed help in data analysis and our first ever multicompartment model based on a beautifully reconstructed entire dendritic branch of a horizontal cell.

NIH funds cell type center

| philippberens

The lab is part of an initiative that NIH will fund to build a BRAIN Initiative comprehensive cell type center to make an atlas of all the cell types in the mouse brain. We will collaborate with the Allen Institute for Brain Science, Baylor College of Medicine, Karolinska, Harvard, Caltech and others in this exciting project! Our team will be responsible for mapping RNAseq-data to physiology, which will provide many new interesting machine learning challenges.    tolias_allcells_big

ERM 2017 in Paris

| philippberens

Image uploaded from iOS (1)Last week, a lab delegation was at the European Retina Meeting 2017 in Paris. It was a fun meeting, with a number of interesting talks and posters on a variety of topics. Two examples include the talk by  Leon Lagnado (also on image) from Sussex who showed that one can infer the release of single vesicles from GluSnFr-experiments. Greg Field from Duke had an interesting talk about coding with DS cells at different light levels. Tübingen was represented strongly with more than 10 posters across labs. Looking forward to the next ERM 2019 in Helsinki and then some work to make ERM 2021 in Tübingen happen!

Joint lab retreat with Euler/Franke labs

| philippberens

We spent two intense days at Kloster Obermarchtal on a joint lab retreat with the labs of Thomas Euler and Katrin Franke to discuss where we are currently at with our projects,  were we want to go and what we want to achieve. Tom Baden, Günther Zeck and Matthias Bethge came as guests to provide feedback and input and discuss collaborations. Also, Tom gave an entertaining talk about the latest on fish vision.

retreat2.jpg

Thanks everyone for the commitment and the discussions!

Lab team in 100k relay race

| philippberens

Together with some people from the Euler lab we participated in the 100k relay race of the University of Tübingen. The race is special because you run in pairs and 1k at a time, ten times each. Effectively, that makes it ten 1k-sprints. We made a very good 11th place out of 37 teams!IMG_0535.JPG

Spikefinder results released!

| philippberens

This week, we have finally released the results of the spikefinder challenge to infer action potentials from calcium recordings. We had more than 50 submissions, many of them in the last few days of the challenge. Obviously, deep learning frameworks were hugely popular, but also “classical” generative modeling approaches (“MLspike”) did remarkably well. Here are the top 10:

top10 submissions.png

We declared Patrick Mineault (Google), Nikolay Chenkov (BCCN Berlin), Peter Rupprecht and Stephan Gerhard (FMI Basel) joint winners, as there submissions were extremely close to each other. A lively debate emerged on twitter about what it all means:

Our preliminary analysis can be found on github. A more complete paper with thorough discussion and analysis will follow.