Biophysically detailed model of CA3 replay

This repository contains a biophysical computational model of a simplified microcircuit of CA3, a hippocampal brain region associated with memory formation. Results obtained with this model are reported in the paper:

Inhibitory plasticity supports replay generalization in the hippocampus. Zhenrui Liao, Satoshi Terada, Ivan Georgiev Raikov, Darian Hadjiabadi, Ivan Soltesz, Attila Losonczy. Nat Neurosci 2024.

During online training, the pyramidal cells in the model receive grid spatially-structured place and grid input and form spatial receptive fields after several simulated laps on a virtual linear track. A fraction of pyramidal cells receive sensory cue input at a randomly selected location during each lap. During the simulated offline period, the pyramidal cells in the model receive random input and undergo epochs of spontaneously sequential activity resembles, consistent with spontaneous memory replay. During these replay-like events, cue cells are suppressed and overall place and cue cell firing is significantly negatively correlated. This model shows that inhibitory plasticity is both sufficient and necessary for cue cell suppression during replay events, and suggests a possible mechanism for spatial map formation that is robust to distractor sensory inputs.

Prerequisites

1. Numpy

The standard python module for matrix and vector computations: https://pypi.python.org/pypi/numpy.

2. Scipy

The standard python module for statistical analysis: http://www.scipy.org/install.html.

3. Matplotlib

The standard python module for data visualization: http://matplotlib.org/users/installing.html.

4. NEURON

A simulator for biophysical models of neurons and networks of neurons: https://github.com/neuronsimulator/nrn

Building Model

Once NEURON is installed, set the PATH and PYTHONPATH environment variables as follows:

export PYTHONPATH=$HOME/install/lib/python:$PYTHONPATH
export PATH=$HOME/install/bin:$PATH

As in a typical NEURON workflow, use nrnivmodl to translate MOD files:

nrnivmodl mechanisms

Running Simulations

Training (online) phase

Run the training (online) phase simulation as follows:

mpirun -n <nprocs> python3 run_training.py \
    --model-home $PWD \
    --circuit-config circuitparams_segs_eegrad_stdp_ee_ie_mf_mec_lec.yaml \
    --arena-config arenaparams_uniform.yaml \
    -c 1229_segs_eegrad_stdp_ee_ie_mf_mec_lec_input_uniform \
    --data-prefix <output_dir> \
    --save-weights-every <n>

where:

• --model-home $PWD : specifies the directory where the model code is located. $PWD indicates the current directory.
• --circuit-config ... : indicates the name of the circuit configuration file. Circuit configuration files are located in subdirectory params by default.
• --arena-config ... : indicates the name of the arena configuration file, which specifies the input stimulus settints. Circuit configuration files are located in subdirectory params by default.
• -c <label> : specifies a label for this model configuration. The label will be used to generate the names of the output files.
• --data-prefix <output-dir> : Optional argument to specify the directory where output files will be written. If not specified, the default is subdirectory data.
• --save-weights-every <n> : Optional argument to specify that weights are saved every n-th lap. If not specified, weights are saved at the end of all laps.

The run_training script generates the following output files:

• cell_spikes_<config name>.npz: spikes produces by the biophysicial neurons in the model.
• ext_spikes_<config name>.npz: spikes produces by the artificial spike sources in the model.
• v_vecs_<config name>.npz: somatic voltage traces of all biophysical neurons in the model.
• <config name>-nlaps-<nlap>.npz: synaptic weights at the end of lap nlap.

Offline phase

Run the offline phase simulation as follows:

mpirun -n <nprocs> python3 run_ripple.py \
        --model-home $PWD \
        --circuit-config circuitparams_ripple.yaml \
        --arena-config arenaparams_ripple_uniform_high.yaml \
        -c ripple_1229_segs_eegrad_stdp_ee_ie_mf_mec_lec_input_uniform \
        --saved-weights-path data/<saved weights file>

where:

• --model-home $PWD : specifies the directory where the model code is located. $PWD indicates the current directory.
• --circuit-config ... : indicates the name of the circuit configuration file. Circuit configuration files are located in subdirectory params by default.
• --arena-config ... : indicates the name of the arena configuration file, which specifies the input stimulus settints. Circuit configuration files are located in subdirectory params by default.
• -c <label> : specifies a label for this model configuration. The label will be used to generate the names of the output files.
• --data-prefix <output-dir> : Optional argument to specify the directory where output files will be written. If not specified, the default is subdirectory data.
• --saved-weights-path <path> : Location of weights file generated by training run.

The run_ripple script generates the following output files:

• cell_spikes_<config name>.npz: spikes produces by the biophysicial neurons in the model.
• ext_spikes_<config name>.npz: spikes produces by the artificial spike sources in the model.
• v_vecs_<config name>.npz: somatic voltage traces of all biophysical neurons in the model.

Model configurations associated with paper

The following model configurations were used to produce the results in the paper. The exact command line invocations used to run simulations for each configuration can be found in directory scripts.

Circuit parameters

• params/circuitparams_segs_eegrad_stdp_ee_ie_mf_mec_lec.yaml : Baseline model configuration
• params/circuitparams_grads_stdp_ee_ie_mf_mec_lec.yaml : Mixed MEC and LEC inputs
• params/circuitparams_segs_eegrad_stdp_ee_mf_mec_lec.yaml : Alternate hypothesis: E->E plasticity
• params/circuitparams_segs_eegrad_stdp_ee_ei_mf_mec_lec.yaml: Alternate hypothesis: E->E and E->I plasticity
• params/circuitparams_segs_eegrad_stdp_ee_ie_mf_mec_lec_ln.yaml: Log-normal distribution of PYR firing rates

Input parameters

• params/arenaparams_uniform.yaml : Input parameters for online phase
• arenaparams_ripple_uniform_high.yaml : Input parameters for offline phase

Analysis

See notebook analysis.ipynb