CLI tools for sequential activity detection in neuronal data which are used for the analysis in the following paper,

Watanabe, K., Haga, T., Euston, D. R., Tatsuno, M., & Fukai, T. (2017). Unsupervised detection of cell-assembly sequences with edit similarity score. bioRxiv, 202655. http:https://doi.org/10.1101/202655

This repo is deprecated. You can find the latest implementation in this repo.

Algorithms are described in bioRxiv paper.

Note that the implementation of clustering algorithms (Optics and Copra) are not included in this repo. You can get the original code from the links shown below:

• Copra: http:https://gregory.org/research/networks/software/copra.html
• OPTICS: https://github.com/espg/OPTICS/blob/master/OPTICS.py

• Python 3
• Julia 0.6
• openjdk "1.8.x"

You need to store your data in data directory. You need two different file for one experiment.

• clu: contain neuron indices. The first row is the number of neurons in it.
• res: contain spike timings unit should be (ms). Note that if the number of data points in res file are $N$, it should be $N+1$ in clu file. And if you have multiple recorded data from multiple shanks, you can process them simultaneously if these file names are defined as following manner,

all.clu.0, all.res.0, all.clu.1, all.res.1, ... all.clu.N, all.res.N

To set PATH to the commands in bin, you need to execute the following line:

source activate

If you want to resume old environment you may run:

source deactivate

or just re-open your terminal.

You can see usage of each command with --help option.

In Chaldea, all processed data will be stored in results. Everytime you start analysis for a novel data set, you need initialization:

init_topdir ../data/tutorial

which will yield root directory (which should be like ../results/tutorial_170202T141337) for successive analysis.

You can add session data with add_session command.

add_session ../data/tutorial/0nneurons_10_seq_duration_100_overlap_0.0 ../results/tutorial_170202T141337

which will create session directory (ex. 0nneurons_10_seq_duration_100_overlap_0.0_170202T152716) which contains the following data.

• activity.npz:
• log.txt:

In Chaldea, binned spike matrix (called binarray) will be used. Following command will create it with user-specified bin width.

 generate_binarray --binwidth 1 ../results/tutorial_170202T141337/0nneurons_10_seq_duration_100_overlap_0.0_170202T152716

Similarity matrix (called simmat. See the article for detail) will be generated by generate_simmat command:

generate_simmat ../results --a 0.05 --p 1 --window 200 --mlenseq 5

Note: this step will take long time.

Options are shown in below.

• a: strength of exponentially growing gap penalty (optional)
• p: number of parallel jobs (optional)
• window: length of sliding time window(ms)
• mlenseq: minimum length of sequence used for calculation time reduction.
• background: run the command in the background
• cluster: run the command in the cluster system

Note: to use cluster option, you need to configure a template job script in bin/cluster directory. A sample script is in there by default.

The simmat has of feature space of sliding time windows that represeant relasionships of each time window. Following clustering enable us to extract time windows that are similar and repetedly occured.

clustering_simmat ../results/tutorial_170202T141337/0nneurons_10_seq_duration_100_overlap_0.0_170202T152716/bin_size1_170202T155547/simmat_window_100a_0.5min_len_3_20170202T195657 --MinPts 5 --v 2

• MinPts: Minimum criteria of cluster. Used in OPTICS.
• v: Maximum number of lables that each data points retain. Used in COPRA. You can make clustering visualization:

visualize_clustering ../results/tutorial_170202T141337/0nneurons_10_seq_duration_100_overlap_0.0_170202T152716/bin_size1_170202T155547/simmat_window_100a_0.5min_len_3_20170202T195657/clusters_MinPts10_v10_170203T100154

Following command will launch small web server to see clustering resutls.

view_clustering_results ../results/tutorial_170202T141337/0nneurons_10_seq_duration_100_overlap_0.0_170202T152716/bin_size1_170202T155547/simmat_window_100a_0.5min_len_3_20170202T195657

Note: Currently, you may have trouble with this command when you use Windows operating system.

From clustering results, common temporal structure in each cluster can be extracted by the following command:

generate_profile ../results/tutorial_170202T141337/0nneurons_10_seq_duration_100_overlap_0.0_170202T152716/bin_size1_170202T155547/simmat_window_100a_0.5min_len_3_20170202T195657/clusters_MinPts8_v3_170203T095951 --p 3 --niter 1000

You can extract sequences by taking common temporal structure between actual spiking activity and the profiles.

extract_sequence ../results/tutorial_170202T141337/0nneurons_10_seq_duration_100_overlap_0.0_170202T152716/bin_size1_170202T155547/simmat_window_100a_0.5min_len_3_20170202T195657/clusters_MinPts8_v3_170203T095951/profiles_numiter_1000_20170203T102604

And you can automatically visualize the results with the following command.

visualize_sequence ../results/tutorial_170202T141337/0nneurons_10_seq_duration_100_overlap_0.0_170202T152716/bin_size1_170202T155547/simmat_window_100a_0.5min_len_3_20170202T195657/clusters_MinPts8_v3_170203T095951/profiles_numiter_1000_20170203T102604/sequences_hosei0.0_20170203T104208/

Also the following command launch a small web browser to check the results.