Detecting complex spikes in extracellular recording

Deep learning-based detection of complex spikes

You can find our preprint article on bioRxiv: https://www.biorxiv.org/content/biorxiv/early/2019/04/05/600536.full.pdf

Tutorial:

A tutorial explaining the workflow and details of the algorithm can be found here: https://colab.research.google.com/github/jobellet/detect_CS/blob/master/Tutorial_detect_cs.ipynb

The tutorial runs in colaboratory: https://research.google.com/colaboratory/faq.html
This enables simple installation because it works online. Also, we can use the GPUs that google provides free of cost.

Implementations:

The software is written in python. You can use the software online using Colaboratory and upload your data in your google drive or use it on your local machine, thanks to a docker container. Both solutions are platform independent and don't require any installation of python.

Colaboratory: the platform independent and online solution

STEP 1: Installing the software (you need to do this step only once)
To install the software, click on this link

This will create a folder named ‘uneye’ in your google drive account where all the scripts and network weights will be stored

STEP 2: Train your network
To train a network, click on this link

This will train a network whose weights can be used later to detect complex spikes. Please look in the tutorial for an example of how to preprocess the data before training.

STEP 3: Detect complex spikes To detect complex spikes in your data, click on this link

Docker: the platform independent and local solution

Disclaimer: this solution might not work with computers with limited computing capacities.

1) Install docker:

for Windows, Mac or Ubuntu

You will need to create an account. For windows user: Once the docker is installed on your machine, you will see the app icon in your task bar. Right click on the icon and select "Switch to Linux containers". You are now ready to type in the terminal.

2) Download the docker image that contains the detect_cs software:

Run the docker app and then type the following command in your terminal::

docker pull joachimbellet/detect_cs

This will download an image containing python and all the dependencies that our algorithm requires (it may take some time).

3) Launch the docker container:

Docker launches a virtual machine that can only access your local files when you instruct it.

Type the following command in your terminal:

docker run -it -p 6888:8888 -v YourPath:/home/jovyan/work joachimbellet/detect_cs

In YourPath write the path on your machine that you want the container to access to. On Unix machine, you might need to write '/' before your path. This folder contains the data you want to analyse.

!! Important: DO NOT copy the output of the command into your browser, but do the following:

Open your web browser. Enter "localhost:6888" as URL. Then you will be asked for a token. This token appeared in your terinal when you entered the command above. It looks something like this: http://6688:8888/?token=775c758f58cdc82bf6ddf51a112228f4dd4229c5b3847bb1 . Copy the token (in this case 775c758f58cdc82bf6ddf51a112228f4dd4229c5b3847bb1) Log in with the token that you saved/copied in the previous step. Now you should see the content of your folder in the web browser. The folder 'work' will contain the path "YourPath" that you entered in the previous step.

4) Start using detect_cs

First use
Run the notebook "Initialize.ipynb". This will create 4 folders in your computer at the path "YourPath":

• "TrainYourNetwork" is where you will drag the files for training your own network
• "Weights" is where the file containing your network weights will be stored after training (you can also drop here the weights of a pretrained network)
• "LabelYourData" is where you will drag the files for which you want CSs to be detected
• "Output" is where the CSs labels will be stored

Training your network

In the folder "TrainYourNetwork", you will drop files containing annotated signals. We recommend to train your network with many different recordings. Thus, in the folder "TrainYourNetwork", it is better to drop many files (>30) containing about 5 examples of complex spikes rather than a few fully-annotated recordings. Each file will be a .mat file and should contain at least 3 variables:

• Either the wide-band or LFP signal
• A high-passed (>600 Hz) filter version of the signal
• A logical array of the same dimension as the signal containing ones for time bins where CS occur and zeros elsewhere.

You can choose the name of your variables. If you have difficulties, see the notebook "Tutorial.ipynb" for an example of how such a file is organized.

Then run the notebook "train_cs_detector.ipynb" and follow the instructions.

This step will generate a trained network whose weights will be saved in the folder "Weights".

Detect your complex spikes

In the folder "LabelYourData" drop your .mat files containing:

• Either the wide-band or LFP signal
• A high-passed (>600 Hz) filter version of the signal You can choose the name of your variables. Then run the notebook "train_cs_detector.ipynb" and follow the instructions.

Then run the notebook "label_your_data.ipynb" and follow the instructions.

The folder Output will then contain the CSs labels corresponding to each file in "LabelYourData"