See the Jupyter Notebooks for the results of EIF extraction for Layer 5 (L5) TTPC2, UTPC and STPC Blue Brain Project (BBP) cells. Data for UTPC is loaded in the /master branch to enable reproducibility of extraction process by using the Notebook. For the complete process of extraction (including the simulation of BBP cell) use the EIF kit and follow the steps below.

EIF extraction steps are replicated from Badel et al. (2008) in which steps are applied to L5 pyramidal neurons and performance of EIF models in capturing the spike timings of real neurons are shown.

This EIF Extraction Kit has 5 files that enables extracting parameters of EIF model (tau, V_resting, V_threshold, delta_t) from a BBP NEURON model.

1. Before starting the simulation, cell should be downloaded, extracted and compiled. NEURON should be downloaded. Follow the steps in https://neuron.yale.edu/neuron/download

2. Download the NEURON cell of your choice from https://bbp.epfl.ch/nmc-portal/downloads.html Remember this kit contains L5 - TTPC1, TTPC2, UTPC, and STPC cells in its dictionary. If you want to investigate another cell, you should add it inside the dataframe "cells_df.pkl"

3. Extract the cell from the .zip file. You can see the possible compilation methods at https://www.neuron.yale.edu/neuron/download/compile_linux on the left of the webpage Windows and Mac compilation methods can be seen. Compilation should be done inside the "mechanisms" file of the cell file. Compiled .mod files should be taken outside the mechanisms file. Placed where the .py files are.

4. Files found in this kit should be placed inside the cell file, together with "run.py" or .mod files that are transferred in the previous step.

5. Open the Simulation_8Types.py. Insert the path of the cell, experiment number and alfa. Run the simulation.

6. Wait for 1-2 hours for the simulation to finish. Have a cup of tea with your family and/or friends.

7. Check the .txt output file

   • Is the firing rate is between 1-15 Hz? If not change the alfa and experiment number, run the simulation again
   • Are the error for the fit and parameters smaller then 0.05? If not, use Recalculate_Parameters.py and calculate again by changing the membrane potential interval. Find the best fit from with smallest error rate and save the graphs and outputs.
   • If you want to see the plots, uncomment the lines for plotting inside EIF_Extraction.py. Also, inside the Recalculate_Parameters.py

Files In This Kit:

*"cells_df.pkl" contains a dataframe where the cell names for the different types and copy numbers are stored. It is necessary for code to find specific cell code given to cell to start working with NEURON. e.g. Layer 5 (L5), Small Tufted Pyramidal Cell (STPC) Copy number 1 has the cell code > cADpyr232_L5_STPC_d16b0be14e

It now has all 5 copies of L5 TTPC1, TTPC2, UTPC and STPC. You can add new cells and copies inside the dataframe by opening in Python.

*OU_process.py is needed to generate Ornstein Uhlenbeck processes for the input.

OU process is a noisy process where noise deviates from mean with standart deviation, but comes back to mean with a time constant tau using the Brownian motion. 2 OU processes with mean 0 and standard deviations [[.18,.18],[.25,.36]] and tau = [3, 10 ms] are used in this simulation following the work of Badel et al., 2008. These processes are combined with 4 different DC components [0, 0.02, 0.03, 0.06]. So, 4 DC components combined with 2 couples of standart deviations, = 8 different OU process are injected to the cell.

No changes will be made to this file.

*"EIF_Extraction.py" has the function that extracts EIF parameters from the given neuron. It follows the steps in the Badel et al., 2008. Simulation injects generated OU process to NEURON cell for 40s after 300 ms transitory time. Injection is to the middle of the soma and outputs obtained from the same spot. Output data is filtered (200 ms after each spike is discarded) and capacitance of cell is found with variance minimization method suggested in the article. Transmembrane current (I_m), Dynamic I-V curve (I_d), and Function of Voltage F(V) is found. F(V) is fitted by EIF model to obtain parameters. Function returns parameters for the EIF model, their error rates, error of the whole fit and saves the output array of simulation for further ivestigation.

One needs to specify the cell type, alfa (amplitude of input, for the firing frequency range of interest), and the trial number to use this function. These parameters are changed inside the Simulation_8Types.py

*Cell type can be changed inside the "Simulation_8Types.py" file. Write the path of the cell by copying from cell folder directory. When directory is changed, code recognizes the name of the cell and copy number. Inputs the name of the cell into EIF_Extraction function.

The directory should be changed manually inside the code in the format: home/user/other folders/layer_cell_type_copy/ e.g. home/ookur/newcells/L5_STPC_cADpyr232_1 It should direct the function inside the cell folder where other .py and .pkl files are.

-Alfa is set to 1. To increase or decrease the firing frequency it can be changed inside the "Simulation_8Types.py" file.

-Trial_no is set to 1. To prevent overwriting or deletion of the output files, it is important to increase the trial number for each experiment inside "Simulation_8Types.py"

*Outputs of the simulation can be observed in .txt file named "output_params.txt"

*I, V, dt, t, and I_m are stored as experiment array (before filtering), and experiment array 2 (after filtering, containing I_m) in .npy files with the name of the cell, copy number and experiment number followed by "_exp_array.npy" and "_exp_array_2.npy" respectively. Experiment array 2 can be used in Recalculate_Parameters.py This will be useful since changing the range of the F(V) values can improve the accuracy of the fit. If the performance of the fit is weak, then this code enables changing the range, seeing the graphs of interest (uncomment to visualize) and obtaining new fits. Best fit can be used for further investigations.

Reference for steps: Badel L, Lefort S, Brette R, Petersen CC, Gerstner W, Richardson MJ. Dynamic I-V curves are reliable predictors of naturalistic pyramidal-neuron voltage traces. J Neurophysiol. 2008 Feb;99(2):656-66. doi: 10.1152/jn.01107.2007. Epub2007 Dec 5. PMID: 18057107.

Enjoy your extraction!

Öykü Okur

For any questions contact > [email protected]