Event-based-neuron

This is the readme for the event-based modeling framework described in the manuscript titled Reproducibility of biophysical in silico neuron states and spikes from event-based partial histories.

To recreate the figures from this paper, begin by running the init.sh script to build the data directories and compile the mod files. This script works for linux and Mac systems, modify this to work in Windows.

The data of this project is generated with the various scripts and script wrappers in ./scripts. The figures themselves are generated within jupyter notebooks within ./notebooks and saved in the figures folder ./figures. All of which utilize code from the ./utils directory.

Python Requirements

• jupyter
• NEURON 8.2
• numpy
• scipy
• matplotlib

Paper Figures

All figures of Cudone et al. 2023 can be reconstructed from the contents of this repository.

• Figure 1: Top and middle generated using notebooks/model_specs.ipynb; which lays out the parameterizations for the Hodgkin-Huxley (HH) and Wang-Buzsaki (WB) point cells for the event-based encodings manuscript. The Wang-Buzsaki is first described in https://doi.org/10.1523/jneurosci.16-20-06402.1996, and built using the implementation from the ModelDB entry 26997.

  Bottom generated using morpho/model_specs_morpho.ipynb; with specifics of the the parameterization for the morphologically detailed CA1 pyramidal cell used in the event-based encodings manuscript. This models uses morphology originally from https://doi.org/10.1002/cne.903620103 and hosted on NeuroMorpho.org (https://doi.org/10.1523/jneurosci.2055-07.2007). The model is based off of Tom Morse's implementation: model 87284 on ModelDB. Our implementation recieves stimuli independently from 15 excitatory and 15 inhibitory synapses at random locations in the cell's dendrites.

• Figure 2: Generated using notebooks/spike_centered.ipynb; Inspired by on conversations with Dr. Robert McDougal and Dr. Ted Carnevale.

• Figure 3: data generated with partial_history.py within the scripts folder for the Hodgkin Huxley models, and partial_history_WB.py within the scripts folder for the Wang Buzsaki models. Figures visualized within notebooks/partial_history.ipynb.

  The partial histories scripts first generate stimuli patterns, then random cell histories (state variable frames) and then run experiments simulating the combination of each of the stimuli patterns with each of the histories with each value of n for n inputs. The results allow for the statistical analysis of response distributions given ambiguity in initial conditions and differing quanta of input stimuli.

• Figures 4: data generated with scripts/nst_distribution_evolution.py and visualized within nst_distribution_evolution_voltage_traces.ipynb.

  Displays an example of how the membranve voltage traces and the distribution of NSTs of a HH model change as more history (input stimuli) is provided, given ambiguity in the initial conditions.

• Figures 5: data generated with scripts/nst_distribution_evolution.py and visualized within nst_distribution_categorical_evolution.ipynb, and nst_distribution_evolution_phase_diagram.ipynb in the notebooks directory.

  nst_distribution_categorical_evolution.ipynb looks at the evolution of the NST distributions as n, the number of stimuli events known by the model, increases. The NST distributions are categorized into one of four categories describing whether the distribution of responses is deterministic or not, and whether it is spiking or not.

  nst_distribution_evolution_phase_diagram.ipynb categorizes the observed response (NST) distributions by whether they are deterministic or not and spiking or not. Then, builds phase diagrams describing how these transitions flow through these categories. Lastly, interogates the specific stimuli conditions (interval between stimuli and type of stimuli) that result in these conditions arising.

• Figures 6: data generated with scripts/nInputs.py for the Hodgkin Huxley model and scripts/nInputs_WB.py for the Wang Buzsaki model, which utilize the wrappers scripts/nInputs_wrapper.sh and scripts/nInputs_WB_wrapper.sh respectively. These scripts run the n inputs experiments which simulates a neuron cell using the on-event framework introduced in Cudone et al. 2023. This requires the mod file nInputs50.mod in the mod folder to be compiled with NEURON.

  Visualization is in notebooks/nInputs_analysis.ipynb, which shows the inter-spike-interval distributions the spike trains of the simulated event-based cells with varying values of n inputs, compared against the standard conductance-based model ISI distributions.

• Figures 7: data generated with scripts/nInputs.py for the Hodgkin Huxley model and scripts/nInputs_WB.py for the Wang Buzsaki model, which utilize the wrappers scripts/nInputs_wrapper.sh and scripts/nInputs_WB_wrapper.sh respectively. These scripts run the n inputs experiments which simulates a neuron cell using the on-event framework introduced in Cudone et al. 2023. This requires the mod file nInputs50.mod in the mod folder to be compiled with NEURON.

  Visualization is in notebooks/nInputs_analysis.ipynb, which shows the victor purpura and van Rossum distances between the spike trains of the simulated event-based cells with varying values of n inputs, compared against the standard conductance-based model.

• Figures 8: nInputs data generated with morpho/scripts/nInputs.py, batch ran using dSQ and the joblist morpho/scripts/nInputs_dSQ_joblist.txt, ground truth simulations of the conductance-based model simulated with morpho/scripts/nInputs_groundtruth_sims.py. Requires stimuli locations files to be generated beforehand using the morpho/scripts/initialize_stimuli_locs.py script.

  Visualized with morpho/nInputs_analysis.ipynb where raster plots of the ground truth conductance-based models are compared to the resulting spike trains of the n Inputs experiments, for the CA1 pyramidal cell. Additionally, van Rossum spike distances measured between the spike trains.

• Figure 9: data for figure 9A, the point cell analysis, is generated from scripts/generate_spiking_histories.py which generates a large sum of spiking histories, and scripts/state_reconstruct.py which runs the pairs of simulations with identical input events but variable initial conditions, and saves thier membrane voltages. The analysis and visualization are within notebooks/state_reconstruction.ipynb, which finds the difference, or error, between the observed membrane voltages, which is the result of the differences in the spiking histories used as initial condtions for the simulation pairs.

  Data for figure 9B, 9B.1 and 9B.2, the morphologically detailed cell analysis, is generated from morpho/scripts/nInputs.py and the ground truth simulation data compared to is generated with morpho/scripts/nInputs_groundtruth_sims.py. The results were run on a cluster using dSQ job query in the morpho/scripts/nInputs_dSQ_joblist.py. The visualization is within morpho/nInputs_analysis.ipynb.

Repository contents

utils

HH.py Contains the HH point cell model as a python class used throughout the codebase.

Morpho.py Contains the morphologically detailed CA1 pyramidal cell as a python class used throughout the codebase. This class uses neuroMorpho's cell morphology file (Ascoli et al. 2007) originally from the work of (Ishizuka et al. 1995). This file is contained in the resources directory in this repository.

Stimuli.py Contains python classes and functions to assist in stimuli generation throughout the codebase; maintains consistency with synaptic parameterizations.

scripts

runs much of the experiments and outputs data to be analyzed in the jupyter notebooks in ./notebooks

data

directory to store intermediate and output data from the python scripts. Internal directory structure for all analysis should be as follows and is instantiated with init.sh.

./nInputs
./nInputs_morpho
./nst_distribution_evolution
    ./results
    ./stimuli
./partial_history
    ./results
./state_reconstruct
    ./original_simulation_data
    ./reconstruction_data
./state_reconstruct_morpho
    ./original_simulation_data

notebooks

holds all of the jupyter notebooks used to generate the figures in the manuscript.

mod

nInputs.mod Custom mod file to integrate event-based partial history input encodings in to NEURON, for the.

morpho_nInputs.mod Custom mod file to integrate event-based partial history input encodings in to NEURON, for the morphologically detailed CA1 pyramidal cell.

Nap.mod Passive Sodium channel used in the dendrites for the CA1 pyramidal cell. Derived from Magistretti & Alonso 1999.

vecevent.mod

Morphology

c91662.CNG.swc The CA1 pyramidal cell morphology file. Downloaded from https://neuromorpho.org/neuron_info.jsp?neuron_name=c91662 (Ascoli et al. 2007). Morphology originally from the work of (Ishizuka et al. 1995)