added run scripts for the various configurations used in the paper

README.md

@@ -2,20 +2,27 @@
 
 This repository contains a biophysical computational model of a
 simplified microcircuit of CA3, a hippocampal brain region associated
-with memory formation. 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.
+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
 
@@ -81,7 +88,7 @@ The run_training script generates the following output files:
 
 ### Offline phase
 
-Run the online phase simulation as follows:
+Run the offline phase simulation as follows:
 
 	mpirun -n <nprocs> python3 run_ripple.py \
             --model-home $PWD \
@@ -106,6 +113,28 @@ The run_ripple script generates the following output files:
 - `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 lines 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](notebooks/analysis.ipynb)

scripts/run_alternate_ee.sh

@@ -0,0 +1,22 @@
+#!/bin/bash
+
+# Alternate hypothesis: E -> E plasticity only
+
+## Online phase (training)
+
+mpirun python3 run_training.py \
+ --model-home $PWD \
+ --circuit-config circuitparams_segs_eegrad_stdp_ee_mf_mec.yaml \
+ --arena-config arenaparams_uniform.yaml \
+ -c 0110_segs_eegrad_stdp_ee_mf_mec_lec_input_uniform \
+ --data-prefix data \
+ --save-weights-every 5
+
+## Offline phase
+
+mpirun python3 run_ripple.py \
+ --model-home $PWD \
+ --circuit-config circuitparams_ripple.yaml \
+ --arena-config arenaparams_ripple_uniform_high.yaml \
+ -c ripple_0110_segs_eegrad_stdp_ee_mf_mec_lec_input_uniform \
+ --saved-weights-path data/0110_segs_eegrad_stdp_ee_mf_mec_lec_input_uniform-cue-ee-ei-nlaps-25-dt-zerodot1-scale-2-v1.npz

scripts/run_alternate_ee_ei.sh

@@ -0,0 +1,22 @@
+#!/bin/bash
+
+# Alternate hypothesis: E -> E and E -> I plasticity only
+
+## Online phase (training)
+
+mpirun python3 run_training.py \
+ --model-home $PWD \
+ --circuit-config circuitparams_segs_eegrad_stdp_ee_ei_mf_mec.yaml \
+ --arena-config arenaparams_uniform.yaml \
+ -c 0110_segs_eegrad_stdp_ee_ei_mf_mec_lec_input_uniform \
+ --data-prefix data \
+ --save-weights-every 5
+
+## Offline phase
+
+mpirun python3 run_ripple.py \
+ --model-home $PWD \
+ --circuit-config circuitparams_ripple.yaml \
+ --arena-config arenaparams_ripple_uniform_high.yaml \
+ -c ripple_0110_segs_eegrad_stdp_ee_ei_mf_mec_lec_input_uniform \
+ --saved-weights-path data/0110_segs_eegrad_stdp_ee_ei_mf_mec_lec_input_uniform-cue-ee-ei-nlaps-25-dt-zerodot1-scale-2-v1.npz

scripts/run_baseline.sh

@@ -0,0 +1,24 @@
+#!/bin/bash
+
+# Baseline model configuration
+
+## Online phase (training)
+
+mpirun 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 0101_segs_eegrad_stdp_ee_ie_mf_mec_lec_input_uniform \
+ --data-prefix data \
+ --save-weights-every 5
+
+## Offline phase
+
+mpirun python3 run_ripple.py \
+ --model-home $PWD \
+ --circuit-config circuitparams_ripple.yaml \
+ --arena-config arenaparams_ripple_uniform_high.yaml \
+ -c ripple_0101_segs_eegrad_stdp_ee_ie_mf_mec_lec_input_uniform \
+ --saved-weights-path data/0101_segs_eegrad_stdp_ee_ie_mf_mec_lec_input_uniform-cue-ee-ei-nlaps-25-dt-zerodot1-scale-2-v1.npz
+
+

scripts/run_lognormal_fr.sh

@@ -0,0 +1,22 @@
+#!/bin/bash
+
+# Log-normal distribution of firing rates configuration
+
+## Online phase (training)
+
+mpirun python3 run_training.py \
+ --model-home $PWD \
+ --circuit-config circuitparams_grads_stdp_ee_ie_mf_mec_lec_ln.yaml \
+ --arena-config arenaparams_uniform.yaml \
+ -c 0117_grads_stdp_ee_ie_mf_mec_lec_ln_input_uniform \
+ --data-prefix data \
+ --save-weights-every 5
+
+## Offline phase
+
+mpirun python3 run_ripple.py \
+ --model-home $PWD \
+ --circuit-config circuitparams_ripple.yaml \
+ --arena-config arenaparams_ripple_uniform_high.yaml \
+ -c ripple_0117_grads_stdp_ee_ie_mf_mec_lec_ln_input_uniform \
+ --saved-weights-path data/0108_grads_stdp_ee_ie_mf_mec_lec_ln_input_uniform-cue-ee-ei-nlaps-25-dt-zerodot1-scale-2-v1.npz

scripts/run_mixed_mec_lec.sh

@@ -0,0 +1,22 @@
+#!/bin/bash
+
+# Mixed MEC/LEC input model configuration
+
+## Online phase (training)
+
+mpirun python3 run_training.py \
+ --model-home $PWD \
+ --circuit-config circuitparams_grads_stdp_ee_ie_mf_mec_lec.yaml \
+ --arena-config arenaparams_uniform.yaml \
+ -c 0108_grads_stdp_ee_ie_mf_mec_lec_input_uniform \
+ --data-prefix data \
+ --save-weights-every 5
+
+## Offline phase
+
+mpirun python3 run_ripple.py \
+ --model-home $PWD \
+ --circuit-config circuitparams_ripple.yaml \
+ --arena-config arenaparams_ripple_uniform_high.yaml \
+ -c ripple_0108_grads_stdp_ee_ie_mf_mec_lec_input_uniform \
+ --saved-weights-path data/0108_grads_stdp_ee_ie_mf_mec_lec_input_uniform-cue-ee-ei-nlaps-25-dt-zerodot1-scale-2-v1.npz

utils/plot_utils.py

@@ -21,10 +21,12 @@ def plot_spikes(spike_times, title, start, finish, gids=None, figsize=(6,2), col
     frs = []
     plt.figure(dpi=600, figsize=figsize)
     i = 0
-
+
+    sts_dict = {}
     for gid in gids:
         sts = spike_times[gid]
         sts = np.asarray(sts) / 1000.
+        sts_dict[gid] = sts
         plt.eventplot(np.asarray(sts).reshape((1,-1)), lineoffsets=i+0.5, 
                       orientation='horizontal',
                       colors=colors)
@@ -40,7 +42,7 @@ def plot_spikes(spike_times, title, start, finish, gids=None, figsize=(6,2), col
     if add_title:
         plt.title('%s fr. mean: %0.3f. std: %0.3f' % (title, np.mean(frs), np.std(frs)))
     plt.show()
-    return
+    return sts_dict
 
 
 def plot_spikes_with_density(spike_times, title, start, finish, gids=None, color='k', binsize=150):
@@ -129,9 +131,9 @@ def plot_avg_peri_event_time_histogram(spike_times, event_time_ranges,
            time_range = time_extent[1] - time_extent[0]
            bin_range = (min(bin_range[0], -time_range/2.), max(bin_range[1], time_range/2))
 
-            
-    avg_rate = np.mean(np.vstack(rates), axis=0)
-    std_rate = np.std(np.vstack(rates), axis=0)
+    rate_array = np.vstack(rates)
+    avg_rate = np.mean(rate_array, axis=0)
+    std_rate = np.std(rate_array, axis=0)
 
     ax.plot(norm_bins, avg_rate,
             color=color,
@@ -143,3 +145,7 @@ def plot_avg_peri_event_time_histogram(spike_times, event_time_ranges,
 
     ax.axvline(0, color='black', linestyle=':')
     ax.set_xlim(bin_range)
+
+    return norm_bins, rate_array
+
+