🧠 Single-Cell-Synaptic-Clustering

This repository contains code and analysis from my Master's thesis project in computational neuroscience. The project investigates how dendritic calcium-mediated action potentials (dCaAPs) influence synaptic clustering in a simulated single cell.

🧩 Project Summary

The simulation includes:

A compartmental neuron with active dendritic compartments
A network of point neurons simulating the input cell assemblies

The goal was to study how dCaAPs enhance dendritic computation and support input clustering and memory stability, in comparison to NMDA spikes.

Key components:

Modeling of somatic and dendritic spiking
Simulating structural and functional plasticity and Spike-Timing Dependent Plasticity (STDP)
Analysis of synaptic input clustering
Comparison of different learning mechanisms

🧪 Simulation Overview

This project simulates a single-compartment neuron connected to 320 point input neurons, divided into 8 assemblies of 40 neurons each. The aim is to explore synaptic clustering under different dendritic integration mechanisms and learning protocols.

We compare two models of the compartmental neuron:

Gidon_mc_lif_group_current.py: Model with dendritic calcium action potentials (dCaAPs)
mc_lif_group.py: NMDA-based model

You can visualize a sample dCaAP spike and its non-monotonic response in the notebook Gidon_dCaAP.ipynb.

Poisson spike trains for input assemblies are generated using poisson_pattern_group.py.

Connections between input neurons and dendrites are set up in layers/rewiring_connection.py, which also implements:

Structural plasticity
Functional plasticity
STDP (Spike-Timing-Dependent Plasticity)
Noise

A schematic of these plasticity mechanisms is provided in Synaptic_dynamics_plots.ipynb.

🧪 Learning Protocols

🔹 Protocol 1: Random Inputs

This protocol tests whether a neuron can learn and retain random, disjoint input assemblies. During each learning window, one of the eight assemblies is randomly selected and activated as a Poisson group firing at 35 Hz, followed by a resting window. This cycle continues for 1000 seconds.

Run: Random.ipynb
Configuration: config_rewiring_ex2.yaml

🔹 Protocol 2: Sequential Inputs

To mimic real-world learning without catastrophic forgetting, assemblies are introduced one-by-one in eight intervals of 125 seconds. During each interval, one assembly fires at 35 Hz while the rest fire at 1 Hz.

Run: Sequential.ipynb
Configuration: config_rewiring_ex3.yaml

🔹 Protocol 3: Overlapping Inputs

Here, assemblies are no longer fully disjoint. Instead, they share overlapping neurons (25%, 50%, or 75%) to simulate shared features between concepts. Assemblies are presented in random order to evaluate learning with shared features.

Run: Overlapping.ipynb
Configuration: config_rewiring_ex5.yaml

🔹 Protocol 4: Coactive Inputs

To simulate simultaneous exposure to multiple stimuli, two or more disjoint assemblies are co-activated randomly. For example, two assemblies fire together for 300 ms, followed by a 200 ms rest, and then a new pair is selected.

Run: Coactive.ipynb
Configuration: config_rewiring_ex4.yaml (set the number of coactive assemblies)

🛠 Running the Simulations

For each protocol, you can choose between the dCaAP model (Gidon_mc_lif_group_current.py) or the NMDA model (mc_lif_group.py). Results will be saved in the results/ folder.

📊 Analyzing Results

Use stats_maker.ipynb to analyze simulation outputs. It generates a stats.txt file in each result directory, summarizing the number of connections and total synaptic weights per dendritic branch.

📄 License and Attribution

All modifications are © 2025 Sima Hashemi and are also licensed under the GNU GPL v3.
See the LICENSE file for full terms.

📖 Citation

If you use this project, please cite both the original work and our modifications.

Original work (on which this project is based):

Limbacher, T., & Legenstein, R. (2020).
Emergence of stable synaptic clusters on dendrites through synaptic rewiring.
Frontiers in Computational Neuroscience, 14, 57.
https://doi.org/10.3389/fncom.2020.00057
GitHub repository

This project:

Hashemi, S., Shafiee, S., & Tetzlaff, C. (2025).
Robust Input Disentanglement Through Calcium-Mediated Dendritic Potentials.
Master's Thesis, Georg-August University of Göttingen.
GitHub repository

simahashemi/Single-Cell-Synaptic-Clustering