NetPyNE

NetPyNE provides a programmatic and graphical interface for the creation, parallel simulation, optimisation, and analysis of data-driven, biophysically detailed multiscale network models in NEURON. This makes multiscale modeling accessible to a wider community, including students and experimentalists; and facilitates and accelerates the workflow of experienced modelers. NetPyNE and the NetPyNE GUI are available via the EBRAINS platform and the Open Source Brain platform. NetPyNE has also been integrated or interfaced with multiple tools in the community including The Virtual Brain (TVB), the Neuroscience Gateway (NSG), Human Neocortical Neurosolver (HNN), SciUnit/SciDash, LFPy and coreNEURON.

Build, simulate and optimise

• Define models using declarative, standardized, rule-based, high-level language (programmatic or via GUI)
• Automatically convert high-level biological specifications to a NEURON network (preventing coding errors and inefficiencies risked with manual implementation)
• Include multiple scales from intra- and extra-cellular molecular reaction-diffusion (RxD) to network level
• Import existing cell models from hoc, Python, NeuroML, and SONATA
• Specify complex connectivity and stimulation patterns at the population, cell, and dendritic level
• Generate a network instance as a Python hierarchical structure with all required NEURON objects
• Easily set up and run parallel simulations, including distributing cells and gathering data from computing nodes
• Run automated parameter explorations/optimizations via grid search or evolutionary algorithms, including simulation submission and management on different environments (local MPI, HPC Slurm/Torque, Neuroscience Gateway)

Visualise and analyse

• Visualize and analyse your model using built-in functions: connectivity, voltage/current traces, molecular concentrations, spike raster plot, firing statistics, information theoretic analysis, etc.
• Position virtual LFP recording electrodes at any arbitrary 3D locations and generate LFP and CSD plots
• Realistic simulation of current dipoles and EEG signals using LFPykit tool
• Load/save from/to common formats (JSON, Pickle, Matlab) and import/export from/to widely-used standards (NeuroML and SONATA)
• Use the web-based GUI to build, run and analyse models, and import/export the model from/to Python scripts

There are 20+ example models available to explore in our GitHub repository and 90+ user-developed models. These include NetPyNE models of many different brain regions (e.g. cortex, thalamus, hippocampus, cerebellum), disease (e.g. ischemia, epilepsy, schizophrenia) and other phenomena (e.g. oscillatory coding, sensorimotor learning, memory retrieval, TMS/tDCS stimulation) [5,6]. Below we provide further details of large-scale thalamocortical models built using NetPyNE. A list of publications making use of NetPyNE is also available.

We used NetPyNE to build and simulate large-scale and biologically realistic models of rodent motor (M1), rodent somatosensory (S1), and macaque auditory (A1) thalamocortical circuits. Each model simulates a cortical column and corresponding thalamic circuits and contains in the order of ~15k detailed neurons ~30M synapses. The model neuronal densities, classes, morphology, biophysics, and connectivity were derived from experimental data. The rodent M1 and S1 models are being used to study neural coding and learning in corticothalamic circuits of the sensorimotor pathway. The macaque A1 model is being used to explore oscillatory mechanisms in the context of speech processing and related disorders. The models may help better understand a variety of diseases, including schizophrenia, Alzherimer’s disease, epilepsy, and motor disorders. See References.

The Potjans-Diesmann cortical microcircuit model is a widely used model originally implemented in NEST. It is a full-scale spiking network model of 1 mm² of cortex, comprising four cortical layers, each containing an excitatory and an inhibitory population, with close to 80,000 point neurons. We reimplemented the model using NetPyNE and reproduced the findings of the original publication. We also implemented a method for scaling the network size that preserves firing rates and other statistics. Our new implementation enabled the use of more detailed neuron models with multicompartmental morphologies and multiple biophysically realistic ion channels. This opens the model to new research, including the study of dendritic processing, the influence of individual channel parameters, the relation to local field potentials (LFPs), and other multiscale interactions.