CNS2023 Workshop (W08):Neuroscience Gateway Enabling Neuroscience Software Dissemination and Large-Scale Neuronal Modeling and Data Processing on Supercomputers


CNS*2023, Leipzig, Germany

Organizers: Amit Majumdar, Subhashini Sivagnanam, Kenneth Yoshimoto

San Diego Supercomputer Center University of California San Diego


Ted Carnevale

Department of Neurobiology Yale University

Workshop day, time: Wednesday, July 19, 09:00 - 12:30 CEST

Abstract of the workshop: The Neuroscience Gateway (NSG- has been enabling large scale modeling and data processing for the neuroscience community for about a decade. It provides about twenty neuroscience software for modeling, data processing and AI/ML work on supercomputers. Examples of these software are NEURON, NEST, NetPyNE, PyNN, BMTK, BluePyOpt, PGENESIS, Brian2, HNN Core, EEGLAB, FreeSurfer, TensorFlow, PyTorch etc. It is free and open to students and researchers from any country. NSG eliminates administrative and technical barriers to using supercomputers for neuroscience research. In recent years, the NSG team acquires close to 30,000,000 core hours per year on academic supercomputers and fairly distributes to NSG users; NSG currently have over 1,580 users. NSG is also a platform for dissemination of neuroscience software. NSG provides a software repository where all the disseminated software is hosted, and a dissemination webpage describing all software disseminated on NSG. NSG's dissemination webpage we provide information about the software, example input file or data, example parameters that need to be provided via NSG's web interface or programmatic interface, and corresponding output files/data.

The workshop speakers are both users of NSG as well as developers of software that are disseminated via NSG. This workshop will provide the perspective of both users and developers of software regarding how NSG is enabling neuroscience research, training and classroom teaching.



Speaker: Subhashini Sivagnanam, San Diego Supercomputer Center, University of California San Diego

Title: Neuroscience Gateway


Description of talk: We will provide a brief overview of the Neuroscience Gateway, including all the tools, software and pipelines that are provided via NSG on supercomputing resources. We will describe how users are using the NSG, developers can use the NSG utilizing the cloud-based developer's platform and NSG enables dissemination of neuroscience software.


Speaker: Robert McDougal, Yale University

Title: Pretraining biophysical models on HPCs using partial history as a path toward larger local models


Description of talk: Neurological diseases remain difficult to treat, due in part to difficulty in translating promising results from animal studies into human treatments. Biophysically-based computational models offer a promising alternative, but are limited due to their computational complexity and to the large number of spatial and temporal scales separating a drug and the desired behavioral outcome. We show that interpreting neurons as event-receiving and sending mechanisms allows the reconstruction of states and spiking behavior of a biophysically-detailed model without the overhead or temporal dependency of integrating large systems of differential equations from initial conditions through continuous time. In this way, using High Performance Computing to build a representation of a cellular model with reduced computational complexity offers the potential to combine realistic responses with larger models for better insight into network outcomes when running on lightweight personal computers.


Speaker: Luca Leonardo Bologna, National Research Council (CNR) - Institute of Biophysics (IBF), Palermo Unit

Title: The EBRAINS Hodgkin-Huxley Neuron Builder: enabling neuron model optimization via the NSG service account


Description of talk: In the framework of the Human Brain Project (HBP) and the EBRAINS Research Infrastructure, we have built the Hodgkin-Huxley Neuron Builder (HHNB), an online resource that allows to: 1) extract electrophysiological features from experimental recordings; 2) choose and/or modify a biophysically and morphologically detailed NEURON model of a single cell; 3) optimize the model parameters against the experimental observations; 4) run the optimized model in a dedicated simulation framework. The core of the HHNB resides in the optimization process that leverages HPC facilities to run the genetic algorithm implemented by the BluePyOpt optimization library. While power users can exploit their own resources on the available HPC systems, this opportunity is not accessible to standard users (e.g., students) with no credentials on supercomputers. To face this issue, in collaboration with the NSG team, we created a service account with the aim of submitting the optimization jobs on behalf of EBRAINS users and give access to a model building and optimization workflow, otherwise difficult to implement. A dedicated service that verifies the user identity, performs the quota management and the job submission, and directly interacts with the supercomputers via the NSG APIs has been developed.

Speaker: Anirban Dutta, University of Lincoln, Brayford Pool, Lincoln, Lincolnshire.

Title: The Open EEGLAB portal to evaluate portable brain imaging of neurovascular coupling in acute brain injury


Description of talk: Searching for cognitive abilities in patients with severe brain injury can be difficult because patients must be sufficiently aroused and able to mobilize motor function to show that they can follow commands. Therefore, the standard neurological assessment often misclassifies clinically unresponsive patients as in a vegetative state (VS, aka. unresponsive wakefulness syndrome, UWS). This has important implications for prognosis and puts unresponsive patients with residual consciousness and cognitive functions at risk of unjustified withdrawal of life-sustaining therapy. Thus, in the past two decades consciousness paradigms based on functional magnetic resonance imaging (fMRI) and electroencephalography (EEG) have been developed that bypass the need for overt motor function. However, although fMRI can detect covert consciousness, fMRI-based paradigms are labour intensive, expensive, logistically challenging and not readily available, while EEG requires electroencephalographer expertise and is subject to artifacts in the intensive care unit. A robust bedside test is clearly needed to assess preserved cognitive abilities in clinically unresponsive patients with brain injury.


Functional near-infrared spectroscopy (fNIRS) is a portable neuroimaging technique that detects simultaneous changes in the optical properties of human cortex due to neurovascular coupling related hemodynamic response to neuronal activation and can be used to display the measures in the form of maps or images of the cortex in disorder of consciousness. We postulated based on our feasibility study that normalization of neurovascular coupling may herald recovery of consciousness after acute brain injury [Othman MH, Bhattacharya M, Moller K, Kjeldsen S, Grand J, Kjaergaard J, Dutta A, Kondziella D. Resting-State NIRS-EEG in Unresponsive Patients with Acute Brain Injury: A Proof-of-Concept Study. Neurocrit Care. 2021 Feb;34(1):31-44. doi: 10.1007/s12028-020-00971-x. PMID: 32333214]. Notably, in a subset of data, post-ictal change in neurovascular coupling from pre-ictal state is a marker of severity. The clinical outcome study is ongoing at the Rigshospitalet, Denmark, led by Dr. Daniel Kondziella's group. The clinical study in neonatal hypoxic ischemic encephalopathy was conducted in India by Dr. Abhijit Das (presently at NHS, UK). The preclinical large animal (ovine) study was conducted by Prof. Satyan Lakshminrusimha's group at the University at Buffalo, USA.


Speaker: Padraig Gleeson, University College London

Title: Integrating the Neuroscience Gateway into Open Source Brain v2.0


Description of talk: Open Source Brain version 1.0 is an online platform for open, collaborative model development in neuroscience. This allows users to visualize, analyse and simulate complex computational models, specified in standardized NeuroML format, through their web browsers. OSBv1 integrated support for NSG's REST interface allowing the models to be automatically sent to the supercomputing resources provided by NSG, simulated, and the results retrieved for analysis, all transparently to the user. Open Source Brain v2.0 is a recent development which extends the platform with connections to data resources such as the Brain Initiative DANDI Archive, to improve the link between models and the experimental data they are constrained by. OSBv2 incorporates a 3D interface for model development and simulation, NetPyNE-UI, and also supports hosted instances of JupyterLab, for interactive data analysis and modelling in Python. We have integrated NSG support into OSBv2 through an updated Python API (, facilitating submission and retrieval of jobs via NSG's REST interface.


Speaker: William Lytton, State University of New York, Downstate HealthSciences University

Title: Making data matter


Description of talk: Many emerging approaches to big data involve the use of machine learning, particularly CNNs, to pull out statistical similarities without concern about meaning or mechanisms. This has become worrisomely apparent with chatGPT, able to achieve enough statistical verisimmilitude to fool some of the people some of the time, without concern for meaning. In neurobiology one can similarly use ML to identify both real and spurious correlations (just as 20 attempts at finding a correlation in rearranged data is likely to pull out one with p<0.05). However, ML can be useful if used in combination with mechanistic mutliscale modeling so as to find method and causality within the statistical fog.