Automated analysis of neural activity

One of the aims of our group is to understand how cognitive processes work at the level of neural circuits (e.g. Tennant et al. 2018; Gerlei et al. 2020 ). Our focus is on mechanisms underlying spatial cognition. Our experimental approach includes recording and manipulating neuronal activity in the brains of rodents performing real world and virtual reality based navigation tasks (Tennant et al. 2018).

We use Eleanor to analyze electrophysiology data generated from these experiments (~5 GB per recording, ~20-40 recordings per animal and ~10 animals per experiment). We integrated Eleanor with our existing Python based analysis pipeline by moving our code onto Eleanor instance images. Our Python code processes raw electrophysiology data to identify the activity of single neurons using MountainSort (Chung et al. 2017), a fully automated spike sorter. Our code then analyzes spatial firing properties of neurons, and visualizes the results for each neuron.

See below an example of our research data and find out more about our analysis pipeline.

Example output of analysis pipeline for a grid cell (Gerlei et al. 2020). The output figures shown here describe the spatial firing properties of a grid cell recorded from a mouse exploring an open field arena. The figures from left to right and top to bottom show the shapes of action potentials on the four channels of the tetrode, autocorrelograms of firing times, number of firing events over time, the speed dependence of firing, a histogram of the position of the animal, firing events overlayed on top of the trajectory of the animal, firing rate map, autocorrelogram of firing rate map, classic head direction plot, firing events on trajectory colour coded for head direction, detected firing fields on the firing rate map and classic head direction plots for each firing field.

Our current workflow allows users to upload their raw data to DataStore and run an automated workflow on Eleanor that outputs the results back to DataStore (Gerlei et al. 2020). To analyze data, we mount the datastore in Eleanor using the Common Internet File System (using mount -t cifs in Linux). We then use Python or rsync to copy data from DataStore and save the results back to DataStore when the analysis is completed. We find this to be more efficient than working on the files in DataStore directly.

Moving to a cloud based system helped us scale up the number of users and also reduced the time we need to spend maintaining local computers. It also helps to keep each user's development environment separate so it is easier to experiment with new approaches. Furthermore, since Eleanor is connected to DataStore we are able to copy the data over quickly compared with connections from a local computer.

Challenges we encountered while setting up analysis on Eleanor include insufficient documentation about connecting the DataStore service with Eleanor, lack of documentation about how to use the openstack API and the lack of tools to estimate and monitor usage cost. To train new users who don’t have a computational background, we wrote step by step instructions that multiple users tested. These instructions are very specific to our use case.

Overall, after the initial efforts of setting up the environment and compiling a detailed user manual specific to our needs, we have been able to use Eleanor successfully to scale up our data analysis. Having a cloud based analysis will enable us to easily increase the amount of data we analyze in the future.

Our work is funded by The Wellcome Trust, The Simons Initiative for the Developing Brain and the Medical Research Council.

Contributors: Klara Gerlei, Teris Tam, Ian Hawes, Matt Nolan.

Why don't you explore 'Research in Action' featured projects demonstrating similar resources and relevant 'Develop your Skills' opportunities? Have a look at the carousels below.