Simulations

Arbor is a portable, high-performance library for computational neuroscience simulations with multi-compartment, morphologically-detailed cells, ranging from single cell models to very large networks. It is written from the ground up with many-CPU and GPU architectures in mind and can be used with contemporary and future HPC systems to meet simulation needs effectively. Arbor separates the neuroscientific description and execution of the simulation, facilitating the sharing and storing of neuroscientific experiments independent from the execution environment. Optimisations make Arbor an order of magnitude faster than the most widely-used comparable simulation software. Arbor can be downloaded as a C++ library and integrated in your own programme, or installed as a Python library (through pip) and imported in any Python scripts.

Snudda is an open-source tool which creates networks of detailed neurons, where the connectivity of the network is based on the morphologies of the neurons. Snudda is specialised on subcortical non-layered structures, and incorporates dynamic neuron modulation. The programme can also be downloaded and run locally on a private workstation, or installed on a supercomputer for generation and simulation of larger networks. Morphological reconstructions are curated, scaled and resampled to ensure the standardised appearance and digital representation in a semi-automatic procedure using assistant Python scripts. Morphologies cut at the slice border are attempted to get repaired. Repaired and normalised morphologies are further validated via HBP Morphology Validation Framework.

NEURON is already well established as a community simulator for networks of detailed neurons. NEURON provides tools for conveniently building, managing, and using models in a way that is numerically sound and computationally efficient. It is particularly well-suited to problems that are closely linked to experimental data, especially those that involve cells with complex anatomical and biophysical properties. CoreNEURON is developed as a new simulation engine executed within the NEURON ecosystem, which seeks to optimise simulation capacity and performance on next-generation hardware. CoreNEURON works by using NEURON to instantiate a model and then remap the data structures to more memory efficient layouts that can leverage modern computing architectures. In addition, models written using the NMODL domain specific language can be better translated to target various backends in order to benefit from the latest compiler optimizations.

The Brain Scaffold Builder is a framework for reconstructing and simulating multi-paradigm neuronal network models. It removes much of the repetitive work associated with writing the required code and lets you focus on the parts that matter. It helps write organized, well-parametrized and explicit code understandable and reusable by your peers. This package is intended to facilitate spatially, topologically and morphologically detailed simulations of brain regions.

This toolset consists of four tools (eFEL, BluePyEfe, BluePyOpt and BluePyMM) that are used sequentially for building and validating biophysically detailed electrical neuron models. The users start by extracting electrical features from the experimental data using the eFEL library by running the BluePyEfe software on their electrophysiological dataset. Next they use the extracted features as constraints for a parameter optimization problem that is solved by the BluePyOpt software. Lastly BluePyMM is used to validate how well the produced electrical models generalize across different neuron morphologies. These software packages are available as open source python code to the community and can easily be installed using the python package installer. They are also integrated in services available on the HBP platform.

Hybrid Molecular Mechanics/Coarse-Grained (MM/CG) simulations help predict ligand poses in human G protein-coupled receptors (hGPCRs), the most important protein superfamily for pharmacological applications. This approach allows the description of the ligand, the binding cavity, and the surrounding water molecules at atomistic resolution, while coarse-graining the rest of the receptor. The Hybrid MM/CG Webserver automatizes and speeds up the MM/CG simulation setup of hGPCR/ligand complexes. Initial structures for such complexes can be easily and efficiently generated with other webservers. The Hybrid MM/CG server also allows for equilibration of the systems, either fully automatically or interactively. The results are visualized online (using both interactive 3D visualizations and analysis plots), helping the user identify possible issues and modify the setup parameters accordingly. Furthermore, the prepared system can be downloaded and the simulation continued locally.

PIPSA (Protein Interaction Property Similarity Analysis) is a method to compare proteins according to their interaction properties. PIPSA may assist in function assignment, and the estimation of binding properties and enzyme kinetic parameters. The PIPSA webserver, webPIPSA, computes protein electrostatic potentials and corresponding similarity indices for a user-defined set of proteins. The standalone code, multipipsa, which includes a python wrapper, provides further options, including running PIPSA on multiple sites on a protein and performing a comparative analysis of the binding properties of user-defined groups of proteins.

SDA (Simulation of Diffusional Association) is a Brownian dynamics simulation software package for the simulation of the diffusion of biomacromolecules in aqueous solution. SDA can be used to compute bimolecular diffusional association rate constants and to predict the structures of diffusional encounter complexes. It can also be used to simulate dilute or concentrated protein solutions and to investigate the adsorption of proteins to solid surfaces. SDA7 is available for standalone use and a subset of the functionality is implemented in the webSDA webserver.

τRAMD is a computationally efficient procedure that enables the computation of relative residence times (τ) or dissociation rates of protein-ligand complexes. It makes use of random acceleration molecular dynamics (RAMD) simulations to facilitate ligand egress in a short timescale and without imposing any bias regarding the ligand egress route. τRAMD is a powerful tool for ranking drug candidates according to their residence times and it can be used with the MD-IFP (Molecular Dynamics - Interaction Fingerprint) tool to investigate dissociation mechanisms and pathways. The combined use of τRAMD and MD-IFP may assist the design of new molecules or ligand optimization.

A novel machine learning method for reweighting free energy profiles and transition state geometries yielded from classical force-field based MD to achieve QM/MM-level accuracy and precision. The method makes use of the targeted reference potential (TRP) method, which maps the MD-derived free energy surface to its quantum mechanical description. It relies on a neural-network (NN) based framework, building on the successful application of novel normalising flows NNs to increase overlap between the configurational distributions associated with the thermodynamic states in free energy perturbation calculations. The TRP method samples configurations from a metadynamics (MD) ensemble, and maps the configurations to “corrected” configurations consistent with those that would be yielded when simulating with a QM/MM potential using a relatively small number of QM/MM energy evaluations as a target. Once fully trained, the neural network maps the cheaper MD derived free energy surface to its more accurate QM/MM counterpart, which can then be used to calculate reliable kinetic binding constants. Additionally, there is the potential to recover transition state configurations at the QM/MM level of theory, which can be analysed further to probe the protein conformational dynamics associated with ligand recognition in neurological receptors. Once at full maturity, this method could be implemented together with metadynamics workflows in computational screening protocols, to predict kinetic binding constants for neurotherapeutic lead compounds of interest to pharmacologists and the wider scientific community.

BioExcel-CV19 is a platform designed to provide web-access to atomistic-molecular dynamics trajectories for macromolecules involved in the COVID-19 disease. The BioExcel-CV19 web server interface presents simulated trajectories, with a set of quality control analyses, system information and interactive and graphical information on key structural and flexibility features. All the analyses integrated in the web portal are completely interactive. Whenever possible, a direct link from the analysis to the 3D representation is offered, using the NGL viewer tool. All data produced is available to download from an associated programmatic access API.

BioExcel Building Blocks Workflows is a collection of biomolecular workflows to explore the flexibility and dynamics of macromolecules, including signal transduction proteins or molecules related to the Central Nervous System. Molecular dynamics setup for protein and protein-ligand complexes are examples of workflows available as Jupyter Notebooks. The workflows are built using the BioBB software library, developed in the framework of the BioExcel Centre of Excellence. BioBBis a collection of Python wrappers on top of popular biomolecular simulation tools, offering a layer of interoperability between the wrapped tools, which make them compatible and prepared to be directly interconnected to build complex biomolecular workflows.

Central Nervous System (CNS) ligands is a platform designed to efficiently generate and parameterize bioactive conformers of ligands binding to neuronal proteins. CNS conformers are generated using a powerful multilevel strategy that combines a low-level (LL) method for sampling the conformational minima and high-level (HL) ab-initio calculations for estimating their relative stability.CNS database presents the results in a graphical user interface, displaying small molecule properties, analyses and generated 3D conformers. All data produced is available to download. CNS ligands provides important data for workflows for parameter generation and mechanistic studies of neuronal cascades using multi-scale molecular simulations in the Human Brain Project.

 

The Glycine Receptor Allosteric Ligand Library (GRALL) is the first database of allosteric modulators of a pharmacologically relevant synaptic receptor. This collection includes agonists, antagonists, positive and negative allosteric modulators of the glycine receptor and a number of experimentally inactive compounds. Importantly, for a large fraction of them a structural annotation based on their putative binding site on the receptor is provided. This type of annotation, which is currently missing in other drug banks, is expected to improve the predictivity of in-silico methodologies for allosteric drug design and boost the development of conformation-based pharmacological approaches.

MoDEL_CNS is a platform designed to provide web-access to atomistic molecular dynamics trajectories for relevant signal transduction proteins. MoDEL_CNS expands the Molecular Dynamics Extended Library (MoDEL) database of atomistic Molecular Dynamics trajectories with proteins involved in Central Nervous System (CNS) processes, including membrane proteins. MoDEL_CNS web server interface presents the resulting trajectories, analyses, and protein properties. All data produced by the project is available to download. MoDEL_CNS will contribute to the improvement of the understanding of neuronal signalling cascades by protein structure-based simulations, calculating molecular flexibility and dynamics, and guiding systems level modelling.

Large scale composite chemical reaction network models are often constructed by combining several smaller models and diverse independent experiments. Constructing a reliable composite model requires a model scoring and optimization framework that can compare the model output to many diverse datasets. FindSim and MOOSE provide a framework for chemical networks simulation (using MOOSE) and comparison to different kinds of experimental datasets (using FindSim). This framework can be used to iteratively optimize model parameters and grow composite models.

The Subcellular Simulation Setup Webapp allows import of SBML model files from the subcellular model building and calibration toolset workflow or other external sources. The tool allows users to setup and configure BioNetGen and STEPS simulations. Users can populate mesh models of spines and other neural structures, and run stochastic simulations of signalling pathways.

The Subcellular Model Building Toolset includes interoperable modules for: model building, calibration (parameter estimation) and model analysis. All information needed to perform these tasks are stored in a structured, human- and machine-readable file format based on SBtab. This information includes: models, experimental calibration data and prior assumptions on parameter distributions. The toolset enables simulations of the same model in simulators with different characteristics, e.g. STEPS, NEURON, MATLAB’s Simbiology and R via automatic code generation.

The SSB Toolkit is a Python library specifically designed for conducting simulations of mathematical models that represent the signal-transduction pathways of G-protein coupled receptors (GPCRs). This library consists of a set of systems biology simulation routines, enabling the investigation of pharmacodynamic models associated with GPCRs. It provides a means to explore how the structural characteristics of these receptors influence subcellular signaling dynamics.

MEDIUM is a Python tool that uses the DF matrices to extract global information on the protein. It works directly on DF images and uses a Convolutional Neural Network (CNN) machine learning approach to train the model in recognising specific patterns in the DF matrix capable of classifying the protein in a specific state in the presence of a ligand (e.g., HOLO or APO states). Then, the trained model can be used to classify the protein state in presence of different ligands, by testing new DF matrices arising from short simulations performed on the new system.

MLCE (Matrix of Lowest Coupling Energies) can be used to represent the region of the protein lowest dynamically coupled. This means that it predicts the region in a protein most likely to be involved in interactions with external partners.

Elephant Visualization (Viziphant) represents a complementary library for Elephant to provide visualisation of data and analysis results. These visualizations assist scientists in quickly interpreting analysis results, as well as helping to keep teaching material, for example the Elephant tutorials, easy to read by removing clutter from the graphics creation.

Neo is a library for handling neurophysiology data in Python. It provides a standardised representation of electrophysiology and optophysiology data in Python, together with support for reading a wide range of neurophysiology file formats, including Spike2, NeuroExplorer, AlphaOmega, Axon, Blackrock, Plexon and Tdt, as well as support for writing to a subset of these formats, plus non-proprietary formats, including NIX and NWB. Neo improves interoperability between Python tools for analysing, visualising and generating electrophysiology data by providing a common, shared object model. Neo objects behave just like normal NumPy arrays, but with additional metadata, checks for dimensional consistency and automatic unit conversion. Neo is a dependency for several other components of EBRAINS, including Elephant, PyNN, Neo Viewer, the Neural Activity Browser and the in-depth data curation workflow. Neo is a community project with contributions from multiple institutions.

The validation test libraries comprise a collection of Python libraries to support structured model validation (NetworkUnit, MorphoUnit, HippoUnit, CerebUnit). All of the libraries are based on the SciUnit framework, developed by the Crook-Gerkin Lab at Arizona State University (members of the HBP Partnering Project “SciUnit”). This framework allows model implementations and validation test implementations to be decoupled by declaring standardised interfaces. This decoupling means that different models of the same structure or phenomenon can be easily validated against the same experimental datasets, and allows us to build up libraries of tests. Our goal is that the majority of models available through KG search will have direct links to the results of automated validation tests, making it easy for EBRAINS users to understand to what extent a given model has been validated. The current tests are grouped into separate Python libraries, according to their domain of application: either the brain region being modelled, or the scope of the model (e.g. single neuron, local network, brain region, etc.).

NeuroScheme is a tool that allows users to navigate through circuit data at different levels of abstraction using schematic representations for a fast and precise interpretation of data. It also allows filtering, sorting and selections at the different levels of abstraction. Finally it can be coupled with realistic visualization or other applications using the ZeroEQ event library developed in WP 7.3.
This application allows analyses based on a side-by-side comparison using its multi-panel views, and it also provides focus-and-context. Its different layouts enable arranging data in different ways: grid, 3D, camera-based, scatterplot-based or circular. It provides editing capabilities, to create a scene from scratch or to modify an existing one.

Visualise neurons and neural circuits consisting of a large number of cells with NeuroTessMesh on your desktop. It enables the visualisation of the 3D morphology of cells included in open databases, such as NeuroMorpho, and provides the tools needed to approximate missing information such as the soma’s morphology. NeuroTessMesh takes morphological tracings of cells acquired by neuroscientists as its only input. It generates 3D models that approximate the neuronal membrane. The resolution of the models can be adapted at the time of visualisation. NeuroTessMesh can assign different colours to different morphologies, in order to visually codify relevant morphological variables, or even neuronal activity.

ViSimpl is a desktop application created to improve your ability to extract knowledge and thereby obtain a better understanding of brain simulation data. It offers the ability to navigate, analyse and interact with brain simulation data, using your choice of either spatial or temporal representations. ViSimpl involves two components: SimPart and StackViz. SimPart is a three-dimensional visualiser data that provides spatio/temporal analysis of the simulation data, using particle-based rendering. StackViz provides a temporal representation of the data at different aggregation levels by illustrating how the electrophysiological variables evolve over time. Together, they allow users to visually discriminate the activity of different groups of neurons, and provide detailed information about individual neurons of interest. These components share synchronised playback control of the simulation being analysed and work together as linked views, although they are loosely coupled and can also be used independently.

Insite is a library that integrates with existing neuronal network simulators and provides a REST API to access the data generated by the simulation while it is running. This includes the generated spikes of the neurons, as well as other measured attributes (e.g. membrane potentials). In addition, it provides access to the network structure, such as the number of neurons and their models and parameters. The framework is meant to be used by third-party visualisation and analysis tools (such as NEST Desktop) to provide a direct view into a running simulation and, for instance, to shorten the time to detect errors. It is the first step towards interactive steering of simulators.

LFPy is an open-source Python module for calculating electric and magnetic brain signals from multi-compartment neuron models and network models. It is built on the NEURON simulator. The different forward models mapping transmembrane currents to the different brain signals in LFPy are incorporated in the simulator-independent package LFPykit, which may also be used for signal predictions with other neural simulators like Arbor.

The HBP Model Validation Service provides web-based tools for working with computational models and validating such models against experimental data. It consists of a REST web service and two clients: the Model Catalog app and a Python client. Underlying metadata are stored in the Knowledge Graph. The service allows users to create, edit, search and view metadata about models and validation tests, and to register, search, view and compare the results of validation tests.

NEST is built for simulations of large networks composed of interconnected simple neuron models. An example of such a model that can be simulated with NEST is the cortical microcircuit. Originally described by Potjans and Diesmann in 2014, the microcircuit model is a building block for larger networks and is of neuroscientific relevance because of the realistic proximity of neurons and synapses. It is a data-driven, full-scale spiking network model of 1 mm² of cortex which relates structure and activity. The model comprises four cortical layers, each containing an excitatory and an inhibitory population, with some 77,000 neurons in total.

A reference implementation is written in PyNEST, the easy Python interface to NEST with commands like Create(), Connect() and Simulate(). As a reference, it has been widely cited and the implementation has been reused in a considerable number of peer-reviewed papers since its original publication.

This is a large-scale spiking model of the vision-related areas of the macaque cortex, including over 4 million neurons (Schmidt et al., 2018). It can be simulated with NEST but requires a larger amount of compute resources. Running the simulation at full scale requires a compute cluster and an advanced workflow to generate model structure from empirical data.

NEST offers the multi-area model and its workflow. Both were implemented following a thorough review process and provide a freely available reference and basis for subsequent work. Watch a video describing the first steps in working with this large multi-scale spiking network model.

Using neural simulators such as NEST, students are able to explore models ranging from single neurons to large neuronal networks. Numerical experiments of increasing complexity help to understand how brain networks function and what the features of their dynamic behaviour are. However, simulators potentially raises the need for knowledge about programming and APIs which tend to be not a major topic in biology or neuroscience classes. Therefore, NEST Desktop has been developed to address this challenge by offering a graphical user interface which enables modelling networks, running simulations with NEST and applying visual data analysis as didactic tool. Its potential has been shown in ongoing courses at Freiburg University on Bachelor and Master level. Didactic concepts and courses are documented as material on NEST Desktop’s documentation page.

Here we explain step by step how to use The Virtual Brain (TVB) tools for end-to-end personalised brain simulation.

The Hodgkin-Huxley Neuron Builder, Single Cell In Silico Experiment Tool (BlueNaaS), Small Circuit In Silico Experiments Tool and Brain Areas Circuit In Silico Experiments Tool allow a multi-scale approach to biophysically detailed neural modelling based on the NEURON simulation environment. At the single level, users can build their own models, optimize and simulate them thanks to the integration with EBRAINS data and tools. At the circuit level, microcircuits of hippocampus or somatosensory cortex neurons can be extracted from the region volumes, to run in-silico experiments. The simulated activity can be downloaded for further analysis. Entire brain regions of the hippocampus can be simulated  too, and the neural activity analysed. These unique tools are not only useful to the research community for the investigation of brain activity at different scales, but they also represent an important support for computational neuroscience educational programs. In addition, we have developed the Hippocampus Hub that provides the scientific community with an easy access to neural data (i.e., electrophysiological recordings, models and morphologies) available on several well-established platforms for neuroscientific resources (e.g., neuromorpho.org, hippocampome.org, kg.ebrains.eu) and allows a user-friendly interaction with the Hodgkin-Huxley Neuron Builder for the construction, optimization and simulation of individual cell NEURON models.

The study by Bruce et al. (2019) and the accompanying live paper provide an example of the concerted use of molecular and subcellular modelling and simulations tools. The study focused on investigating how the regulation of the enzyme adenylyl cyclase 5 in striatal neurons confers the ability to detect coincident neuromodulatory signals. Adenylyl cyclase 5 is regulated by the binding of stimulatory Gαolf and inhibitory Gαi proteins. Protein structure-based molecular dynamics simulations were used to investigate whether Gαolf and  Gαi can bind to Adenylyl cyclase 5  simultaneously and revealed that they could bind and form a stable but inactive ternary complex. The structures of the generated complexes were then used in Brownian dynamics simulations carried out with SDA7 to compute the bimolecular association rate constants for the binding of adenylyl cyclase 5 to the G proteins. The results of these molecular level calculations were then used to constrain a subcellular-level kinetic model that showed that the presence of  this ternary inactive complex is crucial for AC5’s ability to detect  coincident signals, producing a synergistic increase in cAMP. Together with further bioinformatics analysis to compare the structures and sequences of AC isoforms, which included the application of PIPSA (Tong et al., 2016),  the results provided insights into the regulatory mechanisms that control signal processing by different adenylyl cyclase isoforms.

The neuroinformatics platform The Virtual Brain (TVB) offers a modelling framework allowing us to virtually perform stimulation, including DBS, and forecast the outcome from a dynamic systems perspective prior to invasive surgery with DBS lead placement. For an accurate prediction of the effects of DBS, we implement a detailed spiking model of the basal ganglia, which we combine with TVB via our previously developed co-simulation environment (https://arxiv.org/abs/2102.05888). This multiscale co-simulation approach builds on the extensive previous literature of spiking models of the basal ganglia while simultaneously offering a whole-brain perspective on widespread effects of the stimulation going beyond the motor circuit. In the first demonstration of our model, we show that virtual DBS can move the firing rates of a Parkinson’s disease patient’s thalamus - basal ganglia network towards the healthy regime while, at the same time, altering the activity in distributed cortical regions with a pronounced effect in frontal regions. Thus, we provide proof of concept for virtual DBS in a co-simulation environment with TVB. The developed modelling approach has the potential to optimize DBS lead placement and configuration and forecast the success of DBS treatment for individual patients.

Publications: Schirner et al. 2021, Meier et al. 2021; Documentation; Tutorial