Frontiers in Neuroinformatics

A variety of software simulators exist for neuronal networks, and a subset of these tools allow the scientist to model neurons in high morphological detail. The scalability of such simulation tools over a wide range in neuronal networks sizes and cell complexities is predominantly limited by effective allocation of components of such simulations over computational nodes, and the overhead in communication between them. In order to have more scalable simulation software, it is therefore important to develop a robust benchmarking strategy that allows insight into specific computational bottlenecks for models of realistic size and complexity. In this study, we demonstrate the use of the Brain Scaffold Builder (BSB; De Schepper et al., 2021) as a framework for performing such benchmarks. We perform a comparison between the well-known neuromorphological simulator NEURON (Carnevale and Hines, 2006), and Arbor (Abi Akar et al., 2019), a new simulation library developed within the framework of the Human Brain Project. The BSB can construct identical neuromorphological and network setups of highly spatially and biophysically detailed networks for each simulator. This ensures good coverage of feature support in each simulator, and realistic workloads. After validating the outputs of the BSB generated models, we execute the simulations on a variety of hardware configurations consisting of two types of nodes (GPU and CPU). We investigate performance of two different network models, one suited for a single machine, and one for distributed simulation. We investigate performance across different mechanisms, mechanism classes, mechanism combinations, and cell types. Our benchmarks show that, depending on the distribution scheme deployed by Arbor, a speed-up with respect to NEURON of between 60 and 400 can be achieved. Additionally Arbor can be up to two orders of magnitude more energy efficient.

Studying the retina plays a crucial role in understanding how the visual world is translated into the brains language. As a stand-alone neural circuit with easily controllable input, the retina provides a unique opportunity to develop a complete and quantitatively precise model of a computational module in the brain. However, decades of data and models remain fragmented across labs and approaches. To address this, we have launched an open-source retina modelling platform on a shared GitHub repository, aiming to provide a unified data and modelling framework across species, recording techniques, stimulus conditions, and use cases. Our initial release consists of a Python package, openretina, a modelling framework based on PyTorch, which we designed for optimal accessibility and extensibility. The package includes different variations on a basic "Core + Readout" model architecture, easily adaptable dataloaders, integration with modern deep learning libraries, and methods for performing in-silico experiments and analyses on the models. We illustrate the versatility of the package by providing dataloaders and pre-trained models for data from several laboratories and studies across species. With this starter pack in place, openretina can be used within minutes. Through step-by-step examples, we here provide retina researchers of diverse backgrounds a hands-on introduction to modelling, including using models as tools for visualising retinal computations, generating and testing hypotheses, and guiding experimental design.

We present NiftyTorch a Deep Learning Framework for NeuroImaging. The motivation behind the development of such a library is that there are scant amount of centralized tool for deploying 3D deep learning for NeuroImaging. In addition, most of the existing tools require expert technical knowledge in Deep Learning or programming, creating a barrier for entry. The goal is to provide a one stop package using which the users can perform classification tasks, Segmentation tasks and Image Transformation tasks. The intended audience are the members of NeuroImaging who would like to explore deep learning but have no background in coding. In this article we explore the capabilities of the framework, the performance of the framework and the future work for the framework.

FreeSurfer is among the most widely used suites of software for the study of cortical and subcortical brain anatomy. However, analysis using FreeSurfer can be time-consuming and it lacks support for the graphics processing units (GPUs) after the core development team stopped maintaining GPU-accelerated versions due to significant programming cost. As FreeSurfer is a large project with millions of source lines, in this work, we introduce and examine the use of a directive-based framework, OpenACC, in GPU acceleration of FreeSurfer, and we found the OpenACC-based approach significantly reduces programming costs. Moreover, because the overhead incurred by CPU-to-GPU data transfer is the major challenge in delivering GPU-based codes of high performance, we compare two schemes, copy- and-transfer and overlapped-fully-transfer, to reduce such data transfer overhead. Exper-imental results show that the target function we accelerated with overlapped-fully-transfer scheme ran 2.3 as fast as the original CPU-based function, and the GPU-accelerated program achieved an average speedup of 1.2 compared to the original CPU-based program. These results demonstrate the usefulness and potential of utilizing the proposed OpenACC-based approach to integrate GPU support for FreeSurfer which can be easily extended to other computationally expensive functions and modules of FreeSurfer to achieve further speedup.

Reproducible Tract Profiles (RTP) comprises a set of methods to manage and analyze diffusion weighted imaging (DWI) data for reproducible tractography. The tools take MRI data from the scanner and process them through a series of analysis implemented as Docker containers that are integrated into a modern neuroinformatics platform (Flywheel). The platform guarantees that the entire pipeline can be re-executed, using the same data and computational parameters. In this paper, we describe (1) a cloud based neuroinformatics platform, (2) a tool to programmatically access and control the platform from a client, and (3) the DWI analysis tools that are used to identify the positions of 22 tracts and their diffusion profiles. The combination of these three components defines a system that transforms raw data into reproducible tract profiles for publication.\n\nGraphical abstractReproducible Tract Profiles (RTP) comprises a set of methods to manage and analyze diffusion weighted imaging (DWI) data for reproducible tractography. The RTP methods comprise two main parts. O_LIServer side software tools for storing data and metadata and managing containerized computations.\nC_LIO_LIClient side software tools that enable the researcher to read data and metadata and manage server-side computations.\nC_LI\n\nThe server-side computational tools are embedded in containers that are linked to a JSON file with a complete specification of the computational parameters. The data and computational infrastructure on the server is fully reproducible.

Modern electrophysiological experiments in neuroscience typically measure the activity of hundreds of neurons across multiple brain regions. Each recorded neuron has to be assigned its correct area of origin, which can be estimated using histological reconstructions of electrode tracks. However, this approach has a limited accuracy with which the probe position can be determined. While some tools exist to assist in this process in mice, such tools are scarce for other species, such as rats or macaques. Moreover, even in mice, a more reliable brain area assignment may be achieved by using spiking response properties to delineate area boundaries along the recording probe. Here we present the Universal Probe Finder, which provides multi-species support for reconstructing a probes brain area per spike-sorted cluster or recording site. It can use various data formats and can calculate neuronal responsiveness to experimental interventions with only event times and spike times to provide further neurophysiological markers that simplify and improve the reliability of assigning brain areas to neurons. The program can be downloaded here: https://github.com/JorritMontijn/UniversalProbeFinder.

Advancement in modern neuroscience is bottlenecked by neural reconstruction, a process that extracts 3D neuron morphology (typically in tree structures) from image volumes at the scale of hundreds of GBs. We introduce Recut, an automated and accelerated neural reconstruction pipeline, which provides a unified, and domain specific sparse data representation with 79x reduction in the memory footprint. Recuts reconstruction can process 111 Kneurons/day or 79 TB/day on a 24-core workstation, placing the throughput bottleneck back on microscopic imaging time. Recut allows the full brain of a mouse to be processed in memory on a single server, at 89.5x higher throughput over existing I/O-bounded methods. Recut is also the first fully parallelized end-to-end automated reconstruction pipeline for light microscopy, yielding tree morphologies closer to ground truth than the state-of-the-art while removing involved manual steps and disk I/O overheads. We also optimized pipeline stages to linear algorithmic complexity for scalability in dense settings and allow the most timing-critical stages to optionally run on accelerated hardware.

Accurate decoding of neural signals often requires assigning extracellular waveforms acquired on the same electrode to their originating neurons, a process known as spike sorting. While many offline sorters are available, accurate online sorting of spikes with many channels is still a challenging problem. Existing online sorters either use simple algorithms with low accuracy, can only process a handful of channels, or depend on a complex runtime environment that is difficult to set up. We have developed a state-of-the-art online spike sorting platform in Python that enables large-scale, fully automatic real-time spike sorting and decoding on hundreds of channels. Our system is cross-platform and works seamlessly with the Open Ephys suite of open-source hardware and software widely used in many neuroscience laboratories worldwide. It also comes with a user-friendly graphical user interface to monitor the cluster quality, spike waveforms and neuronal firing rate. Our platform has comparable accuracy to offline sorters and can achieve an end-to-end sorting latency of around 160 ms for 128-channel signals. It will be useful for research in fundamental neuroscience, closed-loop feedback neuromodulation and brain-computer interfaces.

Deep neural network models (DNNs) are essential to modern AI and provide powerful models of information processing in biological neural networks. Researchers in both neuroscience and engineering are pursuing a better understanding of the internal representations and operations that undergird the successes and failures of DNNs. Neuroscientists additionally evaluate DNNs as models of brain computation by comparing their internal representations to those found in brains. It is therefore essential to have a method to easily and exhaustively extract and characterize the results of the internal operations of any DNN. Many models are implemented in PyTorch, the leading framework for building DNN models. Here we introduce TorchLens, a new open-source Python package for extracting and characterizing hidden-layer activations in PyTorch models. Uniquely among existing approaches to this problem, TorchLens has the following features: (1) it exhaustively extracts the results of all intermediate operations, not just those associated with PyTorch module objects, yielding a full record of every step in the models computational graph, (2) it provides an intuitive visualization of the models complete computational graph along with metadata about each computational step in a models forward pass for further analysis, (3) it contains a built-in validation procedure to algorithmically verify the accuracy of all saved hidden-layer activations, and (4) the approach it uses can be automatically applied to any PyTorch model with no modifications, including models with conditional (if-then) logic in their forward pass, recurrent models, branching models where layer outputs are fed into multiple subsequent layers in parallel, and models with internally generated tensors (e.g., injections of noise). Furthermore, using TorchLens requires minimal additional code, making it easy to incorporate into existing pipelines for model development and analysis, and useful as a pedagogical aid when teaching deep learning concepts. We hope this contribution will help researchers in AI and neuroscience understand the internal representations of DNNs.

--Neuroimaging study results can vary significantly depending on the datasets and processing pipelines utilized by researchers to run their analyses, contributing to reproducibility issues. These issues are compounded by the fact that there are a large variety of seemingly equivalent tools and methodologies available to researchers for processing neuroimaging data. Here we present NeuroCI, a novel software framework that allows users to evaluate the variability of their results across multiple pipelines and datasets. NeuroCI makes use of Continuous Integration (CI), a software engineering technique, to facilitate the reproducibility of computational experiments by launching a series of automated tests when code or data is added to a repository. However, unlike regular CI services, our CI-based framework uses distributed computation and storage to meet the large memory and storage requirements of neuroimaging pipelines and datasets. Moreover, the frameworks modular design enables it to continuously ingest pipelines and datasets provided by the user, and to compute and visualize results across the multiple different pipelines and datasets. This allows researchers and practitioners to quantify the variability and reliability of results in their domain across a large range of computational methods.

Neuromorphic chips are well-suited for the exploration of neuronal dynamics in (near) real-time. In order to port existing research onto these chips, relevant models of neuronal and synaptic dynamics first need to be supported by their respective development environments and validated against existing simulator backends. At the time of writing, support for short-term synaptic plasticity on neuromorphic hardware is scarce. This technical paper proposes an implementation of dynamic synapses for the SpiNNaker development environment based on the popular synaptic plasticity model by Tsodyks and Markram (TM). This extension is undertaken in the context of existing research on neuromodulation and the study of deep brain stimulation (DBS) effects on singular-neuron responses. The implementation of the TM synapse is first detailed and then, simulated for various response types. Its role in studies of DBS effect on postsynaptic responses is also reviewed. Finally, given the real-time capabilities offered by the hardware, we provide some insight to lay the groundwork for future explorations of closed-loop DBS on neuromorphic chips.

Brightest path tracing is a widely used image processing technique in several fields including biology, geography, and geology. However, despite the availability of many image processing libraries in Python, few offer an out-of-the-box implementation of a bright-est path tracing algorithm. This paper presents a Python package, brightest-path-lib, that efficiently finds the path with maximum brightness between points in a 2D or 3D image. An example graphical user interface is provided as a Napari plugin. Taken together, the package and plugin provide a powerful and extensible tool for users to efficiently trace structures of interest in 2D or 3D images, regardless of the type of structure being analyzed.

Neuromorphic computing currently relies heavily on complicated hardware design to implement asynchronous, parallel and very large-scale brain simulations. This dependency slows down the migration of biological insights into technology. It typically takes several years from idea to finished hardware and once developed the hardware is not broadly available to the community. In this contribution, we present the CloudBrain research platform, an alternative based on modern cloud computing and event stream processing technology. Typical neuromorphic design goals, such as small form factor and low power consumption, are traded for 1) no constraints on the model elements, 2) access to all events and parameters during and after the simulation, 3) online reconfiguration of the network, and 4) real-time simulation. We explain principles for how neuron, synapse and network models can be implemented and we demonstrate that our implementation can be used to control a physical robot in real-time. CloudBrain is open source and can run on commodity hardware or in the cloud, thus providing the community a new platform with a different set of features supporting research into, for example, neuron models, structural plasticity and three-factor learning.

Neural decoding is a powerful method to analyze neural activity. However, the code needed to run a decoding analysis can be complex, which can present a barrier to using the method. In this paper we introduce a package that makes it easy to perform decoding analyses in the R programing language. We describe how the package is designed in a modular fashion which allows researchers to easily implement a range of different analyses. We also discuss how to format data to be able to use the package, and we give two examples of how to use the package to analyze real data. We believe that this package, combined with the rich data analysis ecosystem in R, will make it significantly easier for researchers to create reproducible decoding analyses, which should help increase the pace of neuroscience discoveries.

AO_SCPLOWBSTRACTC_SCPLOWThe analysis of neuronal structure and its relation to function has become a fundamental pillar in neuroscience since its earliest days, with the underlying premise that morphological properties can modulate neuronal computations. It is often the case that the rich three-dimensional structure of neurons is quantified by tools developed in other fields, such as graph theory and computational geometry; nevertheless, some of the more advanced tools developed in these fields have not yet been made accessible to the neuroscience community. Here we present Neural Collision Detection, a library providing high-level interfaces to collision-detection routines and alpha shape calculations, as well as statistical analysis and visualizations for 3D objects, with the aim to lower the entry gap for neuroscientists into these worlds. Our work here also demonstrates a variety of use cases for the library and exemplary analysis and visualizations that were carried out with it on real neuronal and vascular data.

In this work, we developed multiple 2D and 3D segmentation models with multiresolution input to segment brain tumor components, and then ensembled them to obtain robust segmentation maps. This reduced overfitting and resulted in a more generalized model. Multiparametric MR images of 335 subjects from BRATS 2019 challenge were used for training the models. Further, we tested a classical machine learning algorithm (xgboost) with features extracted from the segmentation maps to classify subject survival range. Preliminary results on the BRATS 2019 validation dataset demonstrasted this method can achieve excellent performance with DICE scores of 0.898, 0.784, 0.779 for whole tumor, tumor core and enhancing tumor respectively and accuracy 34.5 % for survuval prediction.

Large-scale diffusion MRI tractography remains a significant challenge. Users must orchestrate a complex sequence of instructions that requires many software packages with complex dependencies and high computational cost. We developed MaPPeRTrac, a probabilistic tractography pipeline that simplifies and vastly accelerates this process on a wide range of high performance computing (HPC) environments. It fully automates the entire tractography pipeline, from management of raw MRI machine data to edge density imaging (EDI) of the structural connectome. Dependencies are containerized with Docker or Singularity and de-coupled from code to enable rapid proto-typing and modification. Data artifacts are strictly organized with the Brain Imaging Data Structure (BIDS) to ensure that they are findable, accessible, interoperable, and reusable following FAIR principles. The pipeline takes full advantage of HPC resources using the Parsl parallel programming frame-work, resulting in the creation of connectome datasets of unprecedented size. MaPPeRTrac is publicly available and tested on commercial and scientific hardware, so that it may accelerate brain connectome research for a broader user community.

The recent development of high-resolution three-dimensional (3D) digital brain atlases and high-throughput brain wide imaging techniques has fueled the generation of large datasets that can be registered to a common reference frame. This registration facilitates integrating data from different sources and resolutions to assemble rich multidimensional datasets. Generating insights from these new types of datasets depends critically on the ability to easily visualize and explore the data in an interactive manner. This is, however, a challenging task. Currently available software is dedicated to single atlases, model species or data types, and generating 3D renderings that merge anatomically registered data from diverse sources requires extensive development and programming skills. To address this challenge, we have developed brainrender: a generic, open-source Python package for simultaneous and interactive visualization of multidimensional datasets registered to brain atlases. Brainrender has been designed to facilitate the creation of complex custom renderings and can be used programmatically or through a graphical user interface. It can easily render different data types in the same visualization, including user-generated data, and enables seamless use of different brain atlases using the same code base. In addition, brainrender generates high-quality visualizations that can be used interactively and exported as high-resolution figures and animated videos. By facilitating the visualization of anatomically registered data, brainrender should accelerate the analysis, interpretation, and dissemination of brain-wide multidimensional data.

Electrophysiology data acquisition of single neurons represents a key factor for the understanding of neuronal dynamics. However, the traditional method to acquire this data is through patch-clamp technology, which presents serious scalability flaws due to its slowness and complexity to record at fine-grained spatial precision (dendrites and axon). In silico biophysical models are therefore created for simulating hundreds of experiments that would be impractical to recreate in vitro. The optimal way to create these models is based on the knowledge of the morphological and electrical features for each neuron. Since large-scale data acquisition is often unfeasible for electrical data, previous expert knowledge can be used but it must have an acceptable degree of similarity with the type of neurons that we are trying to model. Here, we present a data-driven machine learning approach to predict the electrophysiological features of single neurons in case of only having their morphology available. To solve this multi-output regression problem, we use an artificial neural network that has the particularity of providing a probability distribution for every output feature, to incorporate uncertainty. Input data to train the model is obtained from from the Allen Cell Types database. The electrical properties can depend on the morphology, whose acquisition technology is highly automated and scalable so there exist large data sets of them. We also provide integrations with the BluePyOpt library to create a biophysical model using the original morphology and the predicted electrical features. Finally, we connect the resulting biophysical model with the Geppetto UI software to run all the simulations in a sophisticated user interface.

Brain imaging is an increasingly inter-disciplinary field, encompassing multiple data types and multiple analytic traditions. Projects typically involve many moving parts, such as building customized preprocessing pipelines, transforming between data formats, preparing datasets for analysis, and ultimately displaying results. The field is conventionally built on highly specialized software packages that solve these individual challenges well, but are not necessarily designed to be interoperable. Trainees new to the field are therefore often left to come up with isolated heuristics and workarounds to complete a project. Here we present a way to navigate the increasingly complex informatics ecosystem of brain imaging. netneurotools is our labs internal Python toolkit that has been continuously developed and maintained by the labs trainees. The philosophy of the toolkit is that it should be the Swiss army knife of the lab: functions and routines that we often use but that are not part of any established pipeline or package. Since its inception, the toolkit has been open and welcomes contribution from neuroscientists across the globe. netneurotools presents a necessary counterweight to out-of-the-box software packages and highlights the importance of smaller, ad hoc functions for implementing projects. By opening a window into the inner workings of a lab, netneurotools also presents an opportunity to begin a new type of discourse among groups and establish tangible links within the community.