track2p

Cell tracking for longitudinal calcium imaging recordings.

Installation

Setting up the environment

First make sure that you have an updated version of conda installed (the procedure below is based on conda 23.11.0).

Next, set up conda environment with python 3.9:

conda create --name track2p python=3.9
conda activate track2p

Install dependencies:

conda install -c conda-forge matplotlib
conda install -c conda-forge numpy
conda install -c conda-forge scikit-image
pip install itk-elastix

Additionally if using the demo.ipynb notebook you should install jupyter, for example:

conda install conda-forge::jupyterlab

Setting up track2p

For now track2p can only be installed from source via GitHub (TODO: make pip-installable)

For this first clone the repo:

git clone https://github.com/juremaj/track2p

cd to the root directory:

cd track2p

And install the local package (all modules within the track2p/track2p subfolder):

pip install -e .

And thats it, track2p should be succesfully installed :)

Platform specific installation issues/requirements:

MacOS: You might need to have xcode enabled during installation. If you run into xcrun: error: invalid active developer path .../.../... , missing xcrun at .../.../... then enable xcode by running: xcode-select --install

Linux Ubuntu: Installation works, make sure to have Conda up to date, otherwise no issues for now.

Windows: (not tested yet)

Usage

Simple usage

The current version (v1.0.0) is not extremely user friendly, but this will be improved soon. It is recommended to run through run_track2p.py

Recommended: Run via script

This is fairly simple, everything is handled by the run_track2p.py script in the root of the directory. All that needs to be done is to set the paths to all the suite2p folders used for cell tracking by editing the script. The output directory to save the figures and results of tracking should also be defined there. Additional parameters can also be specified by overwriting the defaults (for basic documentation see track2p/ops/default.py, more documentation will be added soon).

After defining the paramters just open a terminal and run:

conda activate track2p
python -m run_track2p

This should start printing out the progress and will tell you once the algorithm is finished :)

Not recommended any more: Jupyter notebook

Another possible way (soon deprecated) is to do it thorugh a jupyter notebook:

conda activate track2p
cd track2p
jupyter lab

And then open demo_t2p.ipynb, where you can then set the paths to the suite2p output directories of the datasets to be matched, as well as the path where the outputs should be saved.

Changing algorithm parameters

TODO: add documentation of algorithm parameters (for now use the defaults saved at track2p/ops/default.py, they should work well)

Outputs

All the outputs of the script will be saved in a track2p folder created within the track_ops.save_path directory specified by the user when running the algorithm. For an introduction on how to use the outputs for further downstream analysis we provide a useful demo notebook demo_t2p_outlput.ipynb in the root of this repository.

Matches

There are two types of output from track2p:

• A matrix (plane#_match_mat.npy) containing the indices of matched neurons across the session for a given plane (# is the index of the plane). Since matching is done from first day to last, some neurons will not be sucessfully tracked after one or a few days. In this case the matrix contains None values. To get neurons tracked across all days only take the rows of the matrices containing no None values.

• A track_ops.npy object that contains the parameters used for the algorithm and some intermediate results that can be used for visualisations (e. g. registered images and ROIs etc.)

Visualisations

There are several visualisations that can be used to evaluate the registration and cell matching quality. These will all be saved in the path defined by track_ops.save_path under track2p/fig

The figures are the following (in the order of importance):

• reg_img_output.png image visualising the quality of image registration across the two days. Each pair of recordings is visualised as red/green overlay of the mean images on the two days before (above) and after (below) registration. If the bottom images don't show good alignment the output of the algorithm would be completely useless (if this happen make sure everyhting is correct with the data)

• thr_met_hist.png visualises the IOU of the matched ROIs. TODO: document how the algorithm works, this will make it clearer. For now refer to other examples for explanation e. g. https://www.cell.com/cell-reports/pdf/S2211-1247(17)31430-4.pdf

• roi_match_plane#_idx###-###.png visualises a window of the mean gcamp image around an example match for all days. If there is more than 100 matches, they are split to separate figures. The dot in the middle shows the centroid of the ROI (whole ROI is not drawn to not bias the estimation of match).

• all_roi_match.png visualises all matches across the FOV for all days. Here the full ROIs are drawn and the color is maintained across days.