CNS 2018 Half-day Workshop

The Neuroscience Gateway and Large Scale Neural Systems Simulations and Tools

Tuesday, 17 July, 2018 | Start time :   2pm
CNS 2018, Seattle, Washington, USA

WORKSHOP ORGANIZERS

Amit Majumdar
San Diego Supercomputer Center and Department of Radiation Medicine and Applied Sciences, 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

Abstract of the workshop:

Large scale modeling and simulations, using supercomputing resources, are important components of computational neuroscience. Computational neuroscientists in the US, from the EU Human Brain Project and those involved with the recently (December, 2017) signed International Brain Initiative (Australia, Japan, Korea, E.U., and US ) depend on High Performance Computing for research. The US NSF and NIH funded Neuroscience Gateway (NSG) project provides neuronal tools, pipelines, and libraries optimally implemented on HPC resources for the neuroscience community. NSG tools and libraries include NEURON, CARLSim, PGENESIS, NEST, Brian, PyNN, MOOSE, BluePyOpt, The Virtual Brain Pipeline, Matlab, EEGLAB, Freesurfer, Human Neocortical Neurosolver etc.; NSG provides tens of millions of supercomputing hours freely for computational neuroscientists, has over 600 users, and is a platform for dissemination of computational neuroscience tools. This workshop will bring together some of the developers of neuronal tools/libraries/pipelines available on NSG and neuroscience users that are using NSG for computational neuroscience research to discuss both tool development and research results enabled by NSG.

Title of Talks and Names of Speakers:

2:00pm - 2:30pm Neuroscience Gateway - Enabling Large Scale Simulations and Data Processing in Neuroscience, Subhashini Sivagnanam, Kenneth Yoshimoto, Amit Majumdar San Diego Supercomputer Center, University of California San Diego, La Jolla, CA USA; Ted Carnevale, Department of Neuroscience, Yale University, New Haven, CT, USA

2:30pm - 3:00pm NeuronUnit: Tools for data-driven validation of neuron and neural circuit models, Richard C. Gerkin, Russell J Jarvis, Sharon M. Crook, Arizona State University, Tempe, AZ, USA

3:00pm - 3:30pm CARLsim 4: An Open Source Library for Large Scale, Biologically Detailed Spiking Neural Network Simulation using Heterogeneous Clusters, Ting-Shuo Chou, Hirak J. Kashyap, Jinwei Xing, Stanislav Listopad, Emily L Rounds, Michael Beyeler, Nikil Dutt, Jeffrey L Krichmar, University of California Irvine, Irvine, CA, USA

3:30pm - 4:00pm Coffee Break

4:00pm - 4:30pm Using NSG to perform millions of simulations in order to characterize in vivo-like states for interneurons of the hippocampus, Alexandre Guet-McCreight, Frances Skinner, Krembil Research Institute, University of Health Network and University of Toronto, Toronto, ON, Canada

4:30pm - 5:00pm Strategies for Parallel NEURON Simulations, Robert McDougal, Department of Neuroscience, Yale University, New Haven, CT, USA

5:00pm - 5:30pm Neuroscience Modeling and Data Processing with Community-authored MATLAB-based Tools, Vijay Iyer, MathWorks Inc., Boston, MA, USA

SCHEDULE OF EVENTS

2pm - 2:30pm 

Neuroscience Gateway - Enabling Large Scale Simulations and Data Processing in Neuroscience

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

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

The Neuroscience Gateway (NSG) is a US National Science Foundation (NSF) and National Institute of Health (NIH) funded project that facilitates access and use of NSF funded High Performance Computing (HPC) resources located at academic supercomputer centers. NSG provides access via a web portal and programmatically to large number of neuroscience tools installed on various HPC resources. . It offers free computer time to neuroscientists acquired via the supercomputer time allocation process managed by the Extreme Science and Engineering Discovery Environment (XSEDE). The NSG provides an administratively and technologically streamlined environment for uploading models, specifying HPC job parameters, querying running job status, receiving job completion notices, and storing and retrieving output data. The NSG transparently distributes user's jobs to appropriate XSEDE HPC resources. NSG is used by both computational neuroscientists and cognitive neuroscientists for large scale simulations and data processing. NSG is also used by developers as a dissemination platform for new neuroscience tools, pipelines and simulation environments. The NSG catalyzes neuroscience research by lowering or eliminating the administrative and technical barriers that currently make it difficult for investigators to use HPC resources and democratizes access to HPC for the broader neuroscience community. In this talk we will discuss the current trend of NSG as it evolves to be a platform for developers, neuroscience educators as well as a collaborative environment for shared access to and processing of data.

2:30pm - 3:00pm 

NeuronUnit: Tools for data-driven validation of neuron and neural circuit models

Richard C. Gerkin, Russell J Jarvis, Sharon M. Crook
Arizona State University, Tempe, Arizona, USA

How do we verify that our models correspond to biology ? Taking a page from the successful software industry practice of "unit-testing", we propose that models should be devel oped with empirical data-driven validation in mind throughout development. When each unit test assesses correspondence w ith one finding from a corresponding biological system, a model's performance on a suite of such tests can serve as an im portant assessment of model validity, and a guide for model development. As barriers to high-performance computing fall, executing such tests continuously during model development, and programmatically optimizing models to pass such tests, b ecomes achievable. We have developed NeuronUnit based on this philosophy. NeuronUnit uses existing community standards to facilitates the construction of tests from a variety of empirical data sources, and offers a model-implementation-inde pendent means for using these tests to assess model fitness. It also offers a web dashboard for comparing diverse models with respect to fitness. We will discuss the use of NeuronUnit and underlying technologies, and invite others to contri bute their ideas and resources for data-driven model validation.

3:00pm - 3:30pm 

CARLsim 4: An Open Source Library for Large Scale, Biologically Detailed Spiking Neural Network Simulation using Heterogeneous Clusters

Ting-Shuo Chou, Hirak J. Kashyap, Jinwei Xing, Stanislav Listopad, Emily L Rounds, Michael Beyeler, Nikil Dutt, Jeffrey L Krichmar
University of California Irvine, Irvine, CA, USA

Large-scale Spiking Neural Network (SNN) simulations are challenging to implement, due to the memory and computation required to iteratively process the large set of neural state dynamics and updates. To meet these challenges, we have developed CARLsim 4, a user-friendly SNN library written in C++ that can simulate large biologically detailed neural networks. Improving on the efficiency and scalability of earlier releases, the present release allows for the simulation using multiple GPUs and multiple CPU cores concurrently in a heterogeneous computing cluster. Benchmarking results demonstrate simulation of 8.6 million neurons and 0.48 billion synapses using 4 GPUs and up to 60x speedup for multi-GPU implementations over a single-threaded CPU implementation, making CARLsim 4 wellsuited for large-scale SNN models in the presence of real-time constraints. Additionally, the present release adds new features, such as Leaky-Integrate-and-Fire (LIF), 9-parameter Izhikevich, multi-compartment neuron models, and fourth order Runge Kutta integration. CARLsim4 is open source and available on the Neuroscience Gateway. 

3:30pm - 4:00pm COFFEE BREAK

4:00pm - 4:30pm 

Using NSG to perform millions of simulations in order to characterize in vivo-like states for interneurons of the hippocampus

Alexandre Guet-McCreight, Frances Skinner
Krembil Research Institute, University of Health Network and University of Toronto, Toronto, ON, Canada

It is technically challenging to obtain recordings from individual cells during behavior, and this is especially true for inhibitory cells - diverse interneuron subtypes that tend to be smaller, less accessible, and less identifiable relative to excitatory cells. However, it is critical to decipher the contributions of interneuron subtypes to achieve an understanding of brain function and behavior (Kepecs & Fishell, 2014). To do this, we are using computational approaches to model inhibitory cells in vivo. To this end, we use Neuroscience Gateway (NSG) resources (Sivagnanam et al, 2013) to do extensive computational explorations of interneuron models in possible in vivo-like scenarios. Our focus is on the hippocampal CA1 interneuron specific 3 (IS3) cell (Tyan et al 2014), a vasoactive intestinal peptide-expressing interneuron subtype that has not yet been recorded from in vivo. Notably, though IS3 cells represent a small fraction of interneurons in CA1 hippocampus, they possess unique circuitry properties in that they only inhibit other inhibitory neurons, and so can function to disinhibit excitatory, pyramidal cells. We have developed data-driven multi-compartment models of IS3 cells with active dendritic properties (Guet-McCreight et al, 2016), determined realistic synaptic parameters along the dendritic morphology of the models (Guet-McCreight et al, 2017), and estimated numbers of active synapses and presynaptic spike rates to generate in vivo-like states for IS3 cell models. To obtain estimates for synaptic inputs in vivo, we explored a full range of synapse numbers, and a wide range of presynaptic spike rates, in two variant IS3 cell models, with and without common inputs. In total this required ~4.4 million 10s multi-compartmental model simulations on the NEURON simulation environment (Version 7.4; Carnevale & Hines, 2006) and analyses. In doing this, we predict that weak and equally balanced synaptic inputs to IS3 cells in the CA1 hippocampus characterize in vivo rhythmic states.

References
Carnevale NT, Hines ML (2006). The NEURON Book. Cambridge, UK: Cambridge University Press.

Guet-McCreight A, et al (2016). eNeuro. 3(4). pii: ENEURO.0087-16.2016.

Guet-McCreight A, et al (2017). F1000Research 2017, 6:1552 (poster).

Kepecs A, Fishell G (2014). Nature. 505(7483):318-26.

Sivagnanam S, et al (2013). IWSG, volume 993 of CEUR Workshop Proceedings, CEUR-WS.org.

Tyan L, et al (2014). J Neurosci. 34(13):4534-47.

4:30pm - 5:00pm 

Strategies for Parallel NEURON Simulations 

Robert McDougal
Department of Neuroscience, Yale University, New Haven, CT, USA

The NEURON simulator and the over 600 published NEURON models available on ModelDB provide a rich starting point for developing, simulating, and analyzing computational models of neuronal phenomena. By implementing NEURON models in a way that supports parallelization, they can take advantage of the many processors available through the Neuroscience Gateway. In NEURON, multiple processors can be used for a single simulation of an individual cell and/or networks; additionally, many simulations can be run simultaneously to study a model's response to parameter changes. We discuss methods enabling each of these parallel strategies, their strengths and limitations, and how to implement models in NEURON that can exploit parallel integration while giving consistent results regardless of the number of processors used. Faster simulation through parallelization makes studying complicated problems more feasible and allows a more natural, rapid exploration of simpler models. 

5:00pm - 5:30pm 

Neuroscience Modeling and Data Processing with Community-authored MATLAB-based Tools 

Vijay Iyer
MathWorks Inc., Boston, MA, USA 

Community-authored freely-shared software tools play a vital role in every area of neuroscience, from the cognitive to the cellular and from the experimental to the theoretical. Many neuroscience-focused tools for modeling neural systems or analyzing neural data are authored in MATLAB, a technical computing environment designed for scientific users. Recently, there are new resources for finding and even running these MATLAB-based community toolboxes for neuroscience. These include the Add-On Explorer, which allows browsing available neuroscience tools by category; and recently the Neuroscience Gateway (NSG), which provides streamlined high-performance computing (HPC) access for neuroscience researchers. NSG now allows its users to run MATLAB code on their HPC resources, including the use of language constructs designed to simplify parallel programming. Additionally, specific tools for EEG data analysis (EEGLAB) and dynamic systems modeling (DynaSim) are pre-installed for common neuroscience applications.