A consumer guide to neuronal modeling software

How Big Data and High-performance Computing Drive Brain Science

Brain science accelerates the study of intelligence and behavior, contributes fundamental insights into human cognition, and offers prospective treatments for brain disease. Faced with the challenges posed by imaging technologies and deep learning computational models, big data and high-performance computing (HPC) play essential roles in studying brain function, brain diseases, and large-scale brain models or connectomes. We review the driving forces behind big data and HPC methods applied to brain science, including deep learning, powerful data analysis capabilities, and computational performance solutions, each of which can be used to improve diagnostic accuracy and research output. This work reinforces predictions that big data and HPC will continue to improve brain science by making ultrahigh-performance analysis possible, by improving data standardization and sharing, and by providing new neuromorphic insights.

Brian 2, an intuitive and efficient neural simulator

Brian 2 allows scientists to simply and efficiently simulate spiking neural network models. These models can feature novel dynamical equations, their interactions with the environment, and experimental protocols. To preserve high performance when defining new models, most simulators offer two options: low-level programming or description languages. The first option requires expertise, is prone to errors, and is problematic for reproducibility. The second option cannot describe all aspects of a computational experiment, such as the potentially complex logic of a stimulation protocol. Brian addresses these issues using runtime code generation. Scientists write code with simple and concise high-level descriptions, and Brian transforms them into efficient low-level code that can run interleaved with their code. We illustrate this with several challenging examples: a plastic model of the pyloric network, a closed-loop sensorimotor model, a programmatic exploration of a neuron model, and an auditory model with real-time input.

Xolotl: An Intuitive and Approachable Neuron and Network Simulator for Research and Teaching

Conductance-based models of neurons are used extensively in computational neuroscience. Working with these models can be challenging due to their high dimensionality and large number of parameters. Here, we present a neuron and network simulator built on a novel automatic type system that binds object-oriented code written in C++ to objects in MATLAB. Our approach builds on the tradition of uniting the speed of languages like C++ with the ease-of-use and feature-set of scientific programming languages like MATLAB. Xolotl allows for the creation and manipulation of hierarchical models with components that are named and searchable, permitting intuitive high-level programmatic control over all parts of the model. The simulator's architecture allows for the interactive manipulation of any parameter in any model, and for visualizing the effects of changing that parameter immediately. Xolotl is fully featured with hundreds of ion channel models from the electrophysiological literature, and can be extended to include arbitrary conductances, synapses, and mechanisms. Several core features like bookmarking of parameters and automatic hashing of source code facilitate reproducible and auditable research. Its ease of use and rich visualization capabilities make it an attractive option in teaching environments. Finally, xolotl is written in a modular fashion, includes detailed tutorials and worked examples, and is freely available at https://github.com/sg-s/xolotl, enabling seamless integration into the workflows of other researchers.

PyMUS: Python-Based Simulation Software for Virtual Experiments on Motor Unit System

We constructed a physiologically plausible computationally efficient model of a motor unit and developed simulation software that allows for integrative investigations of the input-output processing in the motor unit system. The model motor unit was first built by coupling the motoneuron model and muscle unit model to a simplified axon model. To build the motoneuron model, we used a recently reported two-compartment modeling approach that accurately captures the key cell-type-related electrical properties under both passive conditions (somatic input resistance, membrane time constant, and signal attenuation properties between the soma and the dendrites) and active conditions (rheobase current and afterhyperpolarization duration at the soma and plateau behavior at the dendrites). To construct the muscle unit, we used a recently developed muscle modeling approach that reflects the experimentally identified dependencies of muscle activation dynamics on isometric, isokinetic and dynamic variation in muscle length over a full range of stimulation frequencies. Then, we designed the simulation software based on the object-oriented programing paradigm and developed the software using open-source Python language to be fully operational using graphical user interfaces. Using the developed software, separate simulations could be performed for a single motoneuron, muscle unit and motor unit under a wide range of experimental input protocols, and a hierarchical analysis could be performed from a single channel to the entire system behavior. Our model motor unit and simulation software may represent efficient tools not only for researchers studying the neural control of force production from a cellular perspective but also for instructors and students in motor physiology classroom settings.

An action potential-driven model of soleus muscle activation dynamics for locomotor-like movements

OBJECTIVE

The goal of this study was to develop a physiologically plausible, computationally robust model for muscle activation dynamics (A(t)) under physiologically relevant excitation and movement.

APPROACH

The interaction of excitation and movement on A(t) was investigated comparing the force production between a cat soleus muscle and its Hill-type model. For capturing A(t) under excitation and movement variation, a modular modeling framework was proposed comprising of three compartments: (1) spikes-to-[Ca(2+)]; (2) [Ca(2+)]-to-A; and (3) A-to-force transformation. The individual signal transformations were modeled based on physiological factors so that the parameter values could be separately determined for individual modules directly based on experimental data.

MAIN RESULTS

The strong dependency of A(t) on excitation frequency and muscle length was found during both isometric and dynamically-moving contractions. The identified dependencies of A(t) under the static and dynamic conditions could be incorporated in the modular modeling framework by modulating the model parameters as a function of movement input. The new modeling approach was also applicable to cat soleus muscles producing waveforms independent of those used to set the model parameters.

SIGNIFICANCE

This study provides a modeling framework for spike-driven muscle responses during movement, that is suitable not only for insights into molecular mechanisms underlying muscle behaviors but also for large scale simulations.

Why are computational neuroscience and systems biology so separate?

Despite similar computational approaches, there is surprisingly little interaction between the computational neuroscience and the systems biology research communities. In this review I reconstruct the history of the two disciplines and show that this may explain why they grew up apart. The separation is a pity, as both fields can learn quite a bit from each other. Several examples are given, covering sociological, software technical, and methodological aspects. Systems biology is a better organized community which is very effective at sharing resources, while computational neuroscience has more experience in multiscale modeling and the analysis of information processing by biological systems. Finally, I speculate about how the relationship between the two fields may evolve in the near future.

A software for simulating steady-state properties of passive dendrites based on the cable theory

In this study, a computer software, CableTeo, is introduced for simulating the steady-state electrical properties of passive dendrite based on the cable theory. The cable theory for dendritic neurons addresses to current-voltage relations in a continuous passive dendritic tree. It is briefly summarized that the cable theory related to passive cables and dendrites, which is a useful approximation and an important reference for excitable cases. The proposed software can be used to construct user-defined dendritic tree model. The user can define the model in detail, display the constructed dendritic tree, and examine the basic electrical properties of the dendritic tree using transfer impedance approach. The software addresses to ones who want to run simple simulations of the cable theory without need to any programming language skills or expensive software. It can also be used for educational purposes.

A computer software for simulating single-compartmental model of neurons

In this paper, a new computer software package, Yalzer, is introduced for simulating single-compartmental model of neurons. Passive or excitable membranes with voltage-gated ion channels can be modeled, and current clamp and voltage clamp experiments can be simulated. In the Yalzer, first-order differential equations used to define the dynamics of the gate variables and the membrane potential are solved by two separate integration methods with variable time steps: forward Euler and exponential Euler methods. Outputs of the simulation are shown on a spreadsheet template for allowing flexible data manipulation and can be graphically displayed. The user can define the model in detail, and examine the excitability of the model and the dynamics of voltage-gated ion channels. The software package addresses to ones who want to run simple simulations of neurons without need to any programming language skills or expensive software. It can also be used for educational purposes.

Using realistic models to study synaptic integration in cerebellar Purkinje cells

This review presents an approach to modeling which we call "experiments in computo". The use of realistic models makes it possible to generate new predictions that can be confirmed experimentally. Several examples are given of how this approach has improved our understanding of synaptic integration by the Purkinje cell active dendrite. The computer model was constructed to replicate neuronal behavior which has no direct relevance to synaptic integration: it was tuned to reproduce the response of Purkinje cells to current injection in vitro, which consists of a high frequency, regular rhythm of somatic Na+ spikes, interrupted by spontaneous dendritic Ca2+ spikes. The in vivo firing behavior of Purkinje cells is quite different as it consists of highly irregular simple spike firing only, without spontaneous dendritic Ca2+ spikes. The computer model predicted-that the Purkinje cell needs to receive a continuous background inhibitory synaptic drive in addition to the excitation by parallel fibers to obtain this typical in vivo firing. This prediction was confirmed by blocking inhibition during in vivo intracellular recordings. More recently, we demonstrated that the net inhibitory drive to the Purkinje cell dendrite has to be larger than the excitatory synaptic drive. Inhibition hyperpolarizes the dendrite compared to the soma, making it act as a current sink during most of the spiking cycle. These predictions have been confirmed with the dynamic clamp method in the cerebellar slice preparation. Synchronous focal excitatory input by parallel fiber leads in the model to activation of voltage-gated Ca2+ channels which amplify the somatic response by a variable amount. The variability of this graded amplification is due both to position of the input, effectively canceling the cable attenuation, and to the effect of preceding background input. Differences between the model and experimental results in this aspect can be explained by the relative hyperpolarized state of Purkinje cells in the in vitro experimental preparation. These studies led to a new theory about the function of long-term depression in the cerebellum which can explain recent experimental results. In conclusion, our modeling approach generated predictions which contradicted prevalent ideas on how the cerebellum, or neurons in general, works and led to experiments which otherwise would not have been carried out.

Neuromuscular control: introduction and overview

This paper introduces some basic concepts of the interdisciplinary field of neuromuscular control, without the intention to be complete. The complexity and multifaceted nature of neuromuscular control systems is briefly addressed. Principles of stability and planning of motion trajectories are discussed. Closed-loop and open-loop control are considered, together with the inherent stability properties of muscles and the geometrical design of animal bodies. Various modelling approaches, as used by several authors in the Philosophical Transactions of the Royal Society of London, Series B, May 1999 issue, such as inverse and forward dynamics are outlined. An introductory overview is presented of the other contributions in that issue.

A methodology for simulating biological systems using Microsoft Excel

The objective of this present study was to develop a simple, easily understood methodology for solving biologically based models using a Microsoft Excel spreadsheet. The method involves the use of in-cell formulas in which Rows and Columns of new data are generated from data typed into the spreadsheet, but does not require any programming skills or use of the macro language. The approach involves entering the key parameter values into the spreadsheet and conducting the simulation by solving a set of equations based on these parameter values. The examples used in this paper are firstly, a simple voltage clamp simulation in which initial parameter values are used to calculate a system in steady state. The second example is a current clamp simulation where steady state is not reached and the solution of the equations for each time increment is used as the input for the next time increment in the simulation. The calculations are based on the Hodgkin Huxley mathematical equations that describe the voltage dependence of ion channel behavior. The problems and flexibility of the method are briefly discussed. The methodology developed in this present study should help novice modelers to create simple simulations without the need to learn a programming language or purchase expensive software.

A simple analytical method for determining the steady-state potential in models of geometrically complex neurons

A method is presented for solving the cable equation for a spiking neuron below firing threshold or a nonspiking neuron of arbitrary geometry under constant stimulation. The neuron structure is considered as a tree composed of a set of cylinder cables of three types (terminal, intermediate and branching) characterized by their lengths, diameters and linear membrane properties. The stimulation can result from either a uniform conductance-change over a whole cable segment or a point injection of a current. Other special segments are considered (synapses, voltage clamp, lumped soma). Equations are given for replacing any segment with its Thévenin equivalent, i.e. resistance and electromotive force. The step by step use of these elementary equations allows one to find the Thévenin equivalent of the whole neuron and to determine the steady-state membrane potential at any point.

A simple program for simulating the responses of neurons with arbitrarily structured and active dendritic trees

We describe a simple program to simulate neural responses. This program is based on the compartmental approach, in which all compartments of a neuron (i.e. axon, soma or dendrite) are represented by the same basic electrical structure. A parameter file is used to store the model parameters, including the nonlinear channel characteristics of each compartment. The model is then automatically configured according to the values specified in the parameter file. The computation of the conductance of each active channel over time is handled by a unique subroutine optimized according to the kinetics of each channel. The equations for arbitrarily structured trees are solved implicitly using a simple algorithm similar to that of Hines (Hines, M. (1984) Int. J. Biomed. Comput., 15:69-76). The output of the model uses PostScript format. The advantage of this program is that it is small in size, simple to use, efficient, and is platform independent.

Maxsim, software for the analysis of multiple axonal arbors and their simulated activation

In order to analyze the structural organization of complex axonal arbors reconstructed from histological serial sections, and to investigate the functional implications of their geometrical properties, we developed software providing the following facilities: (1) direct importation of data files generated by a commercially available 3-D light-microscopic reconstruction system, including routine procedures for identification and correction of data acquisition errors; (2) real-time 3-D rotations of the arbors in the stack of serial sections; (3) multiple interactive display modes; (4) possibility of modifying diameter and/or connectivity of different branches; (5) simulation of the invasion of the arbor by a single action potential initiated at any chosen point, and visualization of spatio-temporal profiles of activation; (6) extraction of quantitative data converted to standard file formats compatible with available mathematical software. All these tools can be applied to single or multiple axons, individually or simultaneously. The software, called Maxsim, is a highly flexible C-written program running on graphical workstations using the UNIX operating system and X-Window environment.

Computational structure of visual callosal axons

We analysed the activation profiles obtained by simulating invasion of an orthodromic action potential in eleven anterogradely filled and serially reconstructed terminal arbors of callosal axons originating and terminating in areas 17 and 18 of the adult cat. This was done in order to understand how geometry relates to computational properties of axons. In the simulation, conduction from the callosal midline to the first bouton caused activation latencies of 0.9-3.2 ms, compatible with published electrophysiological values. Activation latencies of the total set of terminal boutons varied across arbors between 0.3 and 2.7 ms. Arbors distributed boutons in tangentially segregated terminal columns spanning one or, more often, several layers. Individual columns of one axon were frequently activated synchronously or else with a few hundred microseconds of each other. Synchronous activation of spatially separate columns is achieved by: (i) long primary or secondary branches of similar calibre running nearly parallel to each other for several millimetres; (ii) variations in the calibre of branches serially fed to separate columns by the same primary or secondary branch; (iii) exchange of high-order or preterminal branches across columns. The long, parallel branches blatantly violate principles of axonal economy. Simulated alterations of the axonal arbors indicate that similar spatiotemporal patterns of activity could, in principle, be obtained by less axon-costly architectures. The structure of axonal arbors, therefore, may not be determined solely by the type of spatiotemporal activation profiles it achieves in the cortex but also by other constraints, in particular those imposed by developmental mechanisms.

Modelling the cerebellar Purkinje cell: experiments in computo

Detailed compartmental models of neurons are useful tools for investigating neuronal properties and mechanisms that are not accessible to experimental procedures. If a rigorous approach is used in building the model, simulation studies can be as valuable as laboratory experimentation. As such, modelling becomes an additional method for exploring the function of neurons and nervous systems. As an example, a complex compartmental model with active dendritic membrane of a Purkinje cell is described. The response properties of the model to parallel fiber inputs were investigated. The model fired simple spikes in patterns comparable with those recorded from Purkinje cells in vivo. Synchronous activation of only 20 granule cell inputs was sufficient to generate a measurable response in simulated peri-stimulus histograms. This sensitivity to small excitatory inputs was caused by P-type Ca2+ channels in the dendritic membrane. Such P channels may also be present in the spine heads. Simulations suggest, however, that Ca2+ channels in spine heads cannot be activated by single parallel fiber inputs.

Images of cultured neurons: Morphological and functional information

The in vitro culture of neuronal cells is a useful tool for studying, in a controlled way, neurobiological and neuropharmacological phenomena. A first step towards the understanding of these phenomena is described. Effects of simulated excitatory/inhibitory synapses, artificially positioned along the digitized image of neural arborizations, are presented.

Rallpacks: a set of benchmarks for neuronal simulators

The field of computational neurobiology has advanced to the point where there are several general-purpose simulators to choose from. These cater to various niches in the world of realistic neuronal models, which range from the molecular level to descriptions of entire sensory modalities. In addition, there are numerous custom-designed simulations, adaptations of electrical circuit simulators, and other specific implementations of neurobiological models. As a first step towards evaluating this disparate set of simulators and simulations, and towards establishing standards for comparisons of speed and accuracy, we describe a set of benchmarks. These have been given the name 'Rallpacks' in honor of Wilfrid Rall, who pioneered the study of neuronal systems through analytical and numerical techniques.