Neural circuit regulation of sparsity and pattern separation in the dentate gyrus

This repository contains the code from Galloni, A.R., Samadzelkava, A., Hiremath, K., Oumnov, R., Milstein A.D. (2022). Recurrent Excitatory Feedback From Mossy Cells Enhances Sparsity and Pattern Separation in the Dentate Gyrus via Indirect Feedback Inhibition. Front. in Comp. Neuro., doi:10.3389/fncom.2022.826278

In this project, we construct a series of simple models of the dentate gyrus to explore how different cell types and the distributions of connectivity between them influence population coding and information processing in the network. These are quantified in terms of sparsity, selectivity, and discriminability at the population level within each cell type.

The specifications for each model are stored in yaml files in the "config" directory. These contain the parameters used for each model, including different cell types and their connection properties. Variations of the models provided here can be generated by modifying parameters in these files and rerunning the simulation script (see below).

How to run a single simulation

Requirements

Install requirements by running:

$ pip install -r requirements.txt

Usage

Download this repository by cloning it from github:

$ git clone https://github.com/Milstein-Lab/dentate_circuit_model

A single instance of a model simulation can then be run from the terminal/command line by navigating to the directory where the repository has been saved an executing one of the following commands:

Models in Figure 1 & 2:

• Input-Output with uniform weights:

    $ python simulate_dynamic_model.py --config_file_path=config/simulate_config_0_input_output.yaml --plot
  

• Input-Output with log-normal weights:

    $ python simulate_dynamic_model.py --config_file_path=config/simulate_config_1_input_output.yaml --plot
  

Models in Figure 3:

• FF Inhibition:

    $ python simulate_dynamic_model.py --config_file_path=config/simulate_config_2_FF_Inh.yaml --plot
  

• FF Inhibition, optimized without selectivity constraint:

    $ python simulate_dynamic_model.py --config_file_path=config/simulate_config_2_FF_Inh_no_sel_loss.yaml --plot`
  

Models in Figure 4:

• FB Inhibition:

    $ python simulate_dynamic_model.py --config_file_path=config/simulate_config_3_FB_Inh.yaml --plot
  

• FF + FB Inhibition:

    $ python simulate_dynamic_model.py --config_file_path=config/simulate_config_4_FF+FB_Inh.yaml --plot
  

Models in Figure 5:

• FF Inhibition + indirect FB Inhibition (via excitatory interneurons):

    $ python simulate_dynamic_model.py --config_file_path=config/simulate_config_5_FF_Inh+indirect_FB_Inh.yaml --plot
  

• (-) recurrent connections among the excitatory interneurons:

    $ python simulate_dynamic_model.py --config_file_path=config/simulate_config_5_FF_Inh+indirect_FB_Inh_no_recurrence.yaml --plot
  

• (+) direct FB exitation from Mossy cells:

    $ python simulate_dynamic_model.py --config_file_path=config/simulate_config_6_FF_Inh+indirect_FB_Inh+FB_Exc.yaml --plot
  

Optional flags to add at the end of the command:

• --plot
• --export: generates hdf5 files containing simulation data
• --export_file_name: for saving simulation data to a specific file
• --data_dir: for saving simulation data to a specific directory (by default will save to "/data")

How to optimize parameters and run multiple simulations with parallelization

Requirements

In addition to the requirements above, running simulations with parallelization requires installing:

• MPI for Python

• The nested parallel computing package:

    $ git clone https://github.com/neurosutras/nested.git
  

After cloning nested, add its directory to the PYTHONPATH on your machine

Usage

To run and export data for all simulations used in the paper (with multiple random seeds for each model), execute the following shell script:

$ sh export_model_data.sh config/20211116_model_params.yaml data/exported_dentate_model_data

To optimize parameters for a single model configuration, execute one of the following commands:
(Note this requires significant processing power. Smaller optimizations can be run by reducing the pop_size, max_iter, or path_length)

Models in Figure 1 & 2:

• Input-Output with uniform weights:

    $ mpirun -n 6 python -m mpi4py.futures -m nested.optimize --config-file-path=config/optimize_config_0_input_output_multiple_seeds.yaml --path_length=3 --max_iter=50 --pop_size=200 --disp --framework=mpi
  

• Input-Output with log-normal weights:

    $ mpirun -n 6 python -m mpi4py.futures -m nested.optimize --config-file-path=config/optimize_config_1_input_output_multiple_seeds.yaml --path_length=3 --max_iter=50 --pop_size=200 --disp --framework=mpi
  

Models in Figure 3:

• FF Inhibition:

    $ mpirun -n 6 python -m mpi4py.futures -m nested.optimize --config-file-path=config/optimize_config_2_FF_Inh_multiple_seeds.yaml --path_length=3 --max_iter=50 --pop_size=200 --disp --framework=mpi
  

• FF Inhibition, optimized without selectivity constraint:

    $ mpirun -n 6 python -m mpi4py.futures -m nested.optimize --config-file-path=config/optimize_config_2_FF_Inh_multiple_seeds_no_sel_loss.yaml --path_length=3 --max_iter=50 --pop_size=200 --disp --framework=mpi
  

Models in Figure 4:

• FB Inhibition:

    $ mpirun -n 6 python -m mpi4py.futures -m nested.optimize --config-file-path=config/optimize_config_3_FB_Inh_multiple_seeds.yaml --path_length=3 --max_iter=50 --pop_size=200 --disp --framework=mpi
  

• FF + FB Inhibition:

    $ mpirun -n 6 python -m mpi4py.futures -m nested.optimize --config-file-path=config/optimize_config_4_FF+FB_Inh_multiple_seeds.yaml --path_length=3 --max_iter=50 --pop_size=200 --disp --framework=mpi
  

Models in Figure 5:

• FF Inhibition + indirect FB Inhibition (via excitatory interneurons):

    $ mpirun -n 6 python -m mpi4py.futures -m nested.optimize --config-file-path=config/optimize_config_5_FF_Inh+indirect_FB_Inh_multiple_seeds.yaml --path_length=3 --max_iter=50 --pop_size=200 --disp --framework=mpi
  

• (-) recurrent connections among the excitatory interneurons:

    $ mpirun -n 6 python -m mpi4py.futures -m nested.optimize --config-file-path=config/optimize_config_5c_FF_Inh+indirect_FB_Inh_multiple_seeds.yaml --path_length=3 --max_iter=50 --pop_size=200 --disp --framework=mpi
  

• (+) direct FB exitation from Mossy cells:

    $ mpirun -n 6 python -m mpi4py.futures -m nested.optimize --config-file-path=config/optimize_config_6_FF_Inh+indirect_FB_Inh+FB_Exc_multiple_seeds.yaml --path_length=3 --max_iter=50 --pop_size=200 --disp --framework=mpi