Wscale

Wscale is a specialized tool designed for conducting weight and EPSP (Excitatory Post-Synaptic Potential) grid searches within neural network simulations. This repository offers a robust framework to help researchers and developers optimize neural network parameters, particularly focusing on synaptic weights and EPSP values.

Table of Contents

• Overview
• Features
• Installation
• Usage
  • Grid Search
  • Configuration
  • Simulation
  • Results
• Contributing
• License

Overview

Wscale provides an automated pipeline for performing grid searches over synaptic weights and EPSP values in neural simulations. The tool is highly configurable, allowing users to specify different sections of neurons (e.g., soma, dendrite) and a range of weight values to systematically explore the parameter space.

Features

• Grid Search Capabilities: Perform exhaustive or targeted searches over synaptic weights and EPSP values.
• Flexible Configuration: Define simulation and network parameters through easy-to-edit JSON or Python scripts.
• Modular Design: Easily extend the tool to incorporate new cell types, synaptic mechanisms, or analysis techniques.
• Result Analysis: Calculate EPSP values and loss functions to identify optimal parameter configurations.
• Parallel Processing: Supports parallel computation for speeding up large-scale searches.

Installation

Prerequisites

• Python 3.7+
• pip package manager

Required Libraries

The following Python libraries are required:

• numpy
• scipy
• netpyne
• pandas
• json

Installation Steps

1. Clone the Repository

   git clone https://github.com/yourusername/Wscale.git
   cd Wscale
   

2. Create a Virtual Environment (Optional)

   python3 -m venv venv
   source venv/bin/activate # On Windows use `venv\Scripts\activate`
   

3. Install Dependencies

   pip install -r requirements.txt
   

Usage

Grid Search

To perform a grid search, configure the parameters in the M1_grid_search.py file. Define the sections and weight values you want to explore:

    sections      = ['soma', 'dendrite', 'axon_0', 'axon_1']
    search_params = {
        'sec'    : sections,
        'weight' : [0.001, 0.002, 0.003],
    }

Run the grid search:

    python M1_grid_search.py

Configuration

The configuration of the simulation is handled in the cfg.py file, which defines the key parameters such as simulation duration, timestep, and recording options:

    cfg.duration = 1000    # Simulation duration in ms
    cfg.dt       = 0.025   # Simulation time step in ms
    cfg.verbose  = True    # Verbose output

Simulation

The main simulation setpu is defined in init.py and netParams.py. These scripts initialize network parameters, define cell types, and set up synaptic mechanisms:

    def pt5b_cell():
    PT5B_cell = {
        'secs': {
            'soma': {
                'geom': {
                    'diam': 20.0,   # Diameter of the soma (micrometers)
                    'L': 20.0,      # Length of the soma (micrometers)
                    'Ra': 100.0,    # Axial resistance (ohm*cm)
                    'cm': 1.0       # Membrane capacitance (uF/cm^2)
                },
                'mechs': {
                    'hh': {
                        'gnabar': 0.12,   # Max conductance of sodium channels (S/cm^2)
                        'gkbar': 0.036,   # Max conductance of potassium channels (S/cm^2)
                        'gl': 0.0003,     # Leak conductance (S/cm^2)
                        'el': -54.3       # Leak reversal potential (mV)
                    },
                    'hd': {
                        'gbar': 0.0001    # Max conductance of h-current channels (in S/cm^2)
                    }
                }
            },
        },
    }
    return PT5B_cell

Results

Results from the grid search and simulations are saved to the specified output directory, including computed EPSP values and loss metrics.

    results/
    ├── grid_batch/
    ├── ray/
    └── results.json

These results can be analyzed to identify the optimal synaptic weight and EPSP configurations.