Assemblyfire

Detection of cell assemblies, and analysis of their connectome.

Examples

assemblyfire -v assemblies configs/np_10seeds.yaml  # find assemblies (for every repetition)
assemblyfire -v consensus configs/np_10seeds.yaml  # find "consensus" assemblies (across repetitions)
assemblyfire -v conn-mat configs/np_10seeds.yaml  # extract connectome from SONATA circuit
assemblyfire -v syn-nnd configs/np_10seeds.yaml seed19  # get normalized synapses nearest neighbour distances (in a given repetition)
assemblyfire -v single-cell configs/v7_10seeds.yaml  # calculate spike time reliability (across repetitions)

Once all the features are saved into a single HDF5 file, one can just simply load them, e.g. with the snippet below:

from assemblyfire.config import Config
from assemblyfire.utils import load_assemblies_from_h5
config = Config("configs/np_10seeds.yaml")
assembly_grp_dict, _ = load_assemblies_from_h5(config.h5f_name, config.h5_prefix_assemblies)
for assembly in assembly_grp["seed19"].assemblies:
    assembly_neurons = assembly.gids

or do more involved analyses:

cd analysis_src
python assembly_topology.py  # diverse topological analyses
python compare_assemblies.py  # compare assemblies (within one config, or across configs)
python consensus_botany.py  # features of "consensus" assemblies, and their correlation to spike time reliability
python scan_nclusters.py  # detect different number of clusters of time bins (and corresponding assemblies)

To download the assemblies linked to configs/np_10seeds.yaml and used in our manuscript (see BibTeX below) use the following Zenodo link:

Setting up one's own pipeline

The connectivity analysis part of the package is built to work with the SONATA format (PLoS manuscript, GitHub page) and relies on snap to do so. On the other hand, the cell assembly detection part is based purely on spike times (no information about the connectome is required) and thus can be used as a standalone. To do so one has to set up a new .yaml config specifying at least:

root_path: "<PATH>"

input_sequence_fname: "<FILE_NAME.txt>"

h5_out:
    file_name: "<FILE_NAME>.h5"
    prefixes:
        spikes: "<GROUP_NAME_WITHIN_HDF5_FILE>"
        assemblies: "<GROUP_NAME_WITHIN_HDF5_FILE>"

root_fig_path: "<PATH>"

preprocessing_protocol:
    node_pop: "NAME_OF_NODE_POPULATION_IN_SPIKE_REPORT"  # see below
    target: "NAME_OF_TARGET_IN_SONATA_NODE_SET"  # can be an empty string as well
    t_start:   # in (ms)
    t_end:  # in (ms)
    bin_size:  # in (ms)

Atm. everything is hard coded to our simulation pipeline, thus assemblyfire will look for a pickle file under the root path + analysis/simulations.pkl and will try to load it as a pandas.Series with seed as the name of the index. To change this to your liking, please update utils.py/get_sim_path().

• If the values in the Series are paths to .json files, assemblyfire will assume those to be valid SONATA simulation configs, and try to load them as a bluepysnap.Simulation object, and extract spikes.
• If the values in the Series are paths to .h5 files, assemblyfire will assume those to be valid SONATA spike reports, and try to load them as a libsonata.SpikeReader object, and extract spikes. (For an example for saving spikes in this format, see tests/gen_test_spikes.py).
• If you want to read spikes from an other format, just extend spikes.py/load_spikes(). They should be returned as 2 np.arrays, the first one specifying the times of the spikes, the second one the IDs of the spiking neurons.

Installation

Simply run pip install .

All dependencies are declared in the setup.py and are available from pypi

Citation

If you use this software, kindly use the following BibTeX entry for citation:

@article{Ecker2023,
author = {Ecker, Andr{\'{a}}s and Santander, Daniela Egas and Bola{\~{n}}os-Puchet, Sirio and Isbister, James B. and Reimann, Michael W.},
doi = {https://doi.org/10.1101/2023.02.24.529863},
journal = {bioRxiv},
title = {{Cortical cell assemblies and their underlying connectivity: an in silico study}},
year = {2023}
}

Acknowledgements & Funding

The development of this software was supported by funding to the Blue Brain Project, a research center of the École polytechnique fédérale de Lausanne (EPFL), from the Swiss government’s ETH Board of the Swiss Federal Institutes of Technology.

Copyright (c) 2023 Blue Brain Project / EPFL.