deep learning in the lab

Technology advances in neuroscience on all fronts, from new microscopy to chemical and molecular techniques to recording devices. We use and develop some of these tools but in this post I want to highlight how we adapted tools generated by @trackingactions and @deeplabcut in the mouse motor lab. Nick did the legwork to setup deeplabcut so we could use a machine learning approach to tracking the behavior of rats in our social choice tests.

To validate the approach we conducted simple social novelty preference tests which typically reveal a rat's tendency to spend more time investigating a new social interaction partner rather than a familiar one. This graphic is an overlay of nose positions tracked from 11 rats for every frame (30Hz) of video in a 5 minute social preference test. Red areas indicate the most time, can you tell where the test rat preferred? The dashed box on the right! This is the location of a chamber that contained a novel social partner and the left box contained a cagemate. We are now using this approach to integrate in vivo fiber photometry measures of insular cortex activity in social affective preference tests. Take a look at Nick's report about this at biorxiv.