Neuroscience Gateway (NSG) 

NSG and HPAC – Large Scale Simulations and Data Processing

 

Saturday 19 October, 2019 | 08:30 AM - 12:30 PM
Chicago Downtown, USA
Location: Will be announced to attendees

Register by October 10, 2019 at https://neuron.yale.edu/neuron/static/courses/nsg2019/nsg2019.html

 

WORKSHOP ORGANIZERS

Amit Majumdar
San Diego Supercomputer Center, University of California San Diego, La Jolla, CA, USA

Subhashini Sivagnanam
San Diego Supercomputer Center, University of California San Diego, La Jolla, CA, USA

Ted Carnevale
Department of Neuroscience, Yale University, New Haven, CT, USA

Alexander Peyser
Jülich Supercomputing Center, Forschungszentrum Jülich, Germany

Abstract of the workshop:

The Neuroscience Gateway Portal (NSG https://www.nsgportal.org/) enables neuroscientists to do large scale simulations and data processing on the US NSF funded supercomputers and academic cloud resources located at various academic supercomputing centers. It eliminates administrative and technical barriers by providing free supercomputing time to users and easy access to widely used neuroscience tools that currently includes BluePyOpt, Brian, BMTK, CARLsim, DynaSim, EEGLAB, Freesurfer, GENESIS, Human Neocortical Neurosolver (HNN), MATLAB, MOOSE, NEST, NetPyNE, NEURON, Octave, PyNN, Python, R, TensorFlow etc. NSG's web-based interface simplifies the tasks of uploading models or data, specifying job parameters, monitoring job status, and storing and retrieving output data. NSG allows tool developers to disseminate their newly developed tools on NSG which has a growing user base. NSG is also used for teaching at classrooms and workshops. The Human Brain Project's High Performance Analytics and Computing Platform (HPAC https://hbp-hpc-platform.fz-juelich.de/) provides extensive HPC support for the HBP community. This includes developing and providing the hardware and software infrastructure required for large scale simulations, data management and analysis, visualization, and managing the complex workflows involved in performing these tasks. Speakers at the workshop will discuss their research where large scale modeling, data processing and data management are important.

Title of Talks and Names of Speakers:

08:30 – 08:45    Introduction of the workshop.

08:45 – 09:15 Introduction to the Neuroscience Gateway and High Performance Analytics and Computing Platform, Amit Majumdar, Subhashini Sivagnanam, San Diego Supercomputer Center, UCSD, La Jolla, CA, USA ; Ted Carnevale, Department of Neuroscience, Yale University, New Haven, CT, USA; Alexander Peyser, Jülich Supercomputing Center, Forschungszentrum Jülich, Germany

09:15 – 09:45 How to create artificial spike train populations as inputs for detailed single neuron simulations, Samira Abbasi and Dieter Jaeger, Emory University, Atlanta, GA, USA

09:45 – 10:15 Probing Flexibility of the Predictive Brain Using Computation-Intensive Approaches and EEG, Seydanur Tikir, Albert Einstein College of Medicine, Bronx, NY, USA

10:15 – 10:45 Studying Deep Brain Simulation Using Calcium Imaging via a Head-Mounted Miniature Microscope, J. K. Trevathan1, A. J. Asp2, E. N. Nicolai1,2, J. M. Trevathan2, N.A. Kremer3, J. L. Lujan3,4, K. A. Ludwig1,5; 1Department of Bioengineering, University of Wisconsin, Madison, WI 53706, USA, 2Mayo Clinic Graduate School of Biomedical Sciences, Rochester, MN 55905, USA,

3Department of Neurologic Surgery, Mayo Clinic, Rochester, MN 55905, USA, 4Department of Physiology and Biomedical Engineering, Mayo Clinic, Rochester, MN 55905, USA, 5Department of Neurological Surgery, University of Wisconsin, Madison, WI 53706, USA

10:45 – 11:00 BREAK

11:00 – 11:30 A large-scale neuronal network model of the trisynaptic pathway of rat hippocampus, Gene J. Yu, University of Southern California, Los Angeles, CA, USA

11:30 – 12:00 Construction and Applications of Biologically-Realistic Multi-Scale Models of the Mouse Primary Visual Cortex, Anton Arkhipov, Allen Institute for Brain Science, Seattle, WA, USA

12:00 – 12:30 LUNCH (provided)

 

SCHEDULE OF EVENTS

08:45 - 09:15 

Introduction to the Neuroscience Gateway

Amit Majumdar, Subhashini Sivagnanam
San Diego Supercomputer Center, University of California San Diego, La Jolla, CA, USA

Ted Carnevale
Department of Neuroscience, Yale University, New Haven, CT, USA

Alexander Peyser
Jülich Supercomputing Center, Forschungszentrum Jülich, Germany

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, and how developers interact with the NSG and the NSG team to implement their tools, software and pipelines. We will describe the various research and development activities ongoing at the High Performance Analytics and Computing Platform of the Human Brain Project.

09:15 - 09:45 

How to create artificial spike train populations as inputs for detailed single neuron simulations.

Samira Abbasi and Dieter Jaeger
Emory University, Atlanta, Georgia, USA

Realistic synaptic input patterns that correctly reflect the conditions in behaving animals are an important component of exploring biophysical neuron model performance.  I am presenting a general method to construct such input patterns as clones from recorded spike trains, but with the ability to generate arbitrary correlation strength and shared rate changes within populations of such cloned spike trains while maintaining their in vivo inter-spike interval statistics.

09:45 – 10:15

Probing Flexibility of the Predictive Brain Using Computation-Intensive Approaches and EEG

Seydanur Tikir

Albert Einstein College of Medicine, Bronx, NY, USA

The brain actively produces predictions of upcoming events in everyday life and update these predictions according to new information. We test the hypothesis that individuals with autism spectrum disorders (ASD) do not flexibly adjust their predictions following changes in environmental statistics, with the use of electroencephalography (EEG), behavior, and modeling. We will present our recent data showing how the amplitudes of certain evoked potentials (i.e., P3, CNV) are gradually modulated by environmental statistics in the neurotypical group, and how this pattern is different in the ASD group. This work required significant computational resources due to the use of high-density EEG (160 scalp electrodes), a high sampling rate (every two milliseconds), and several hours of collected data. We leveraged key advances in NSG to facilitate EEG analysis while employing computation-intensive algorithms. The talk will incorporate tactics for managing complexity in EEG analysis, modular processing of data, and automating the pipeline while ensuring methodological quality.

 

10:15 – 10:45

Studying Deep Brain Stimulation Using Calcium Imaging via a Head-Mounted Miniature Microscope

J. K. Trevathan1, A. J. Asp2, E. N. Nicolai1,2, J. M. Trevathan2, N.A. Kremer3, J. L. Lujan3,4, K. A. Ludwig1,5

1Department of Bioengineering, University of Wisconsin, Madison, WI 53706, USA

2Mayo Clinic Graduate School of Biomedical Sciences, Rochester, MN 55905, USA

3Department of Neurologic Surgery, Mayo Clinic, Rochester, MN 55905, USA

4Department of Physiology and Biomedical Engineering, Mayo Clinic, Rochester, MN 55905, USA

5Department of Neurological Surgery, University of Wisconsin, Madison, WI 53706, USA

After decades of study in humans and animal models, there remains a lack of consensus regarding how the action of electrical stimulation leads to the therapeutic effects of neuromodulation. To further our understanding, there is a critical need for the application of novel methodological approaches using state-of-the-art neuroscience tools to the study of neuromodulation in preclinical models of disease. We are addressing this need by showing technical feasibility and demonstrating utility of one approach combining chronic behaving single-photon microendoscope recordings in a 6-OHDA mouse model of Parkinson’s disease (PD) with electrical stimulation of a common deep brain stimulation (DBS) target, the subthalamic nucleus (STN).

The experimental protocol to obtain calcium imaging in a behaving 6-OHDA lesioned mouse consisted of, induction of calcium indicator expression via a unilateral injection of an AAV9 viral vector into striatum, placement of a Gradient Refractive Index (GRIN) lens into the target recording location just above striatum, induction of a PD phenotype via unilateral injection of 6-OHDA into the substantia nigra pars compacta (SNc), and placement of a bipolar stimulating electrode within the STN for application of electrical stimulation. Neural activity recorded during a series of stimulation parameters applied in anesthetized and behaving mice were processed on Neuroscience Gateway using constrained nonnegative matrix factorization for microendoscope data (CNMF-E) run in Matlab 2016b. Neural activity traces extracted using CNMF-E were used to assess the effects of STN DBS. Our data demonstrate that calcium imaging during neuromodulation therapies enables measurement of stimulation frequency-dependent changes in neural activity that differed between anesthetized and awake conditions. This work represents a first step towards using calcium imaging to study the effects of stimulation on deep brain neuronal activity in an animal model of PD and is an example of how Neuroscience Gateway can be used to accelerate the analysis of large calcium imaging data sets.

10:45 - 11:00 BREAK

11:00 – 11:30

A large-scale neuronal network model of the trisynaptic pathway of rat hippocampus

Gene J. Yu

University of Southern California, Los Angeles, CA, USA

The axonal projections between neural populations are often anatomically organized resulting in ordered topographic maps that determine the connectivity of the neural system. They are seldom, if at all, connected uniformly random or all-to-all. The hippocampus is a system that clearly demonstrates a higher-level organization as described by the trisynaptic circuit in which activity predominantly propagates in a feedforward manner through the three hippocampal subfields. In addition, each projection exhibits its own distinct topography. However, there have been few studies that investigate how the individual topographies affect the dynamics of each subfield and the activity transformations that each subfield performs.

We have developed a computational platform for simulating the full-scale, in terms of numbers of neurons and synapses and the geometric volume, of a rat hippocampus. Using this platform, we have constructed a neuronal network using the NEURON simulation environment which incorporates the unique topographies of the perforant path projection, the mossy fiber projection, the CA3 associational system, and the Schaffer collaterals. In this work, we created a network composed of 112 000 entorhinal cortical cells, 1 200 000 granule cells, 250 000 CA3 pyramidal cells, and 380 000 CA1 pyramidal cells that include basket cells for each subfield. The geometric area for the axon terminal fields for each projection were constrained based on anatomical studies of the respective axonal distributions. Topography was ultimately found to determine the spatial extent of the clusters, predominantly along the longitudinal axis of the hippocampus. This was further quantified as the longitudinal distance at which neuronal spiking could be correlated.

The results of this work demonstrate a clear role for topography in determining the spatio-temporal dynamics of a neural system. The work also establishes a neuronal network model that can be used to investigate the trisynaptic pathway at a scale that encompasses the entire rat hippocampus.

11:30 - 12:00

Construction and Applications of Biologically-Realistic Multi-Scale Models of the Mouse Primary Visual Cortex

Anton Arkhipov

Allen Institute for Brain Science, Seattle, WA, USA

The systematic experimental platforms at the Allen Institute for Brain Science characterize cell types, connectivity, and neural activity in the mouse visual cortex. By carefully integrating these data, we have constructed highly realistic 230,000-neuron models of the mouse cortical area V1, receiving thalamocortical visual inputs, at the biophysically detailed and point-neuron levels of resolution.  Applications include systematic comparisons of simulated responses to in vivo experiments, investigations of the structure-function relationships in cortical circuits, and simulated optogenetic perturbations, all of which result in experimentally testable predictions. This work is enabled by our modeling software suite called Brain Modeling ToolKit (BMTK) and the SONATA file format.  These tools and the models are all shared freely with the community.

 

12:00 – 12:30 LUNCH (provided)