Data Contribution and Sharing
Contribution of Neuron Image Data Sets
The set of BigNeuron bench-test data will include neuron image stacks from different species (including fruit fly and other insects, fish, turtle, chicken, mouse, rat, and human) and nervous system regions such as cortical and subcortical areas, retina, and peripheral nervous system. Many of these neurons come from large-scale neuroinformatics projects such as the Allen Mouse and Human Cell Types projects, Taiwan FlyCircuits, and Janelia Fly Light, but a number of data sets are also contributed directly by neuroscientists worldwide. Several data sets have also been reconstructed using alternative methods and some reconstructions had been manually curated and/or proofread.
The test data will be from multiple light microscopy modalities, especially laser scanning microscopy (confocal/2p) and brightfield or epi-fluorescent imaging. The neurons are labeled using different methods, such as genetic labeling and virus/dye/biocytin injection, and will span a broad range of types (e.g. unipolar, multipolar, releasing different neurotransmitters, and with a wide variety of electrophysiological properties).
The current edition of BigNeuron will consider only 3D image stacks of single neurons or neurons that have relatively clear separation in their arborization patterns. Neurons labeled using Brainbow-type techniques will be included if suitable for reliable color separation processing. Focusing the large-scale bench-testing on 3D single-neuron image stacks will maximize feasibility. Future editions of BigNeuron would consider other challenges such as separating neurons in densely labeled samples, resolving connectivity, time-lapse tracing, electron microscopy data, etc.
The contributed data sets will be preprocessed to ensure usability for bench-test (see "Data Protocol") on supercomputers. These data will be archived and will remain available for future bench-test. We encourage neurobiologists to contribute as many appropriate neuron data sets as possible. We also encourage such data sets to be manually reconstructed as much as possible, thus to provide a suitable “gold standard” for evaluation of automated neuron reconstruction methods. The data contributors will enter a royalty-free Open Data agreement (see "Open Data"). Importantly, the data contributors will also receive a royalty-free copy of all reconstructions, thus saving a substantial amount of resource to quantify the morphology of their neurons.
We expect BigNeuron bench-test will generate a large number of reconstructions corresponding to different algorithms, roughly 10-20, for each neuron image data set. The consensus reconstruction, as well as alternative ways to look at the population of such reconstructions such as principal components of reconstructions, will be produced as well. All these data will be shared in a primary new public web-based database. The conditions associated with the generation of such data will also be documented and shared in the database.
BigNeuron reconstructions will be analyzed as a community effort. The project will provide user-friendly data visualization and analysis tools (see "Data Visualization and Analysis Protocol"), as well as guidance to adapt previously published and available methods such as L-Measure, DIADEM scores, and spatial distances. However, the specific analyses will be determined as a collaborative effort by any and all interested research groups. The research community will jointly define the most appropriate approaches to quantify the data, thus more objectively defining the state of the art of single neuron reconstruction.
We will also encourage the community to contribute manually created or curated reconstructions to be used as references or “gold standards” to evaluate the automatically produced reconstructions. This will also help characterize the types of errors in the reconstructions produced by the automated algorithms.
- Interested people who would like to contribute neuron image data sets should register by sending an email email@example.com.
- Each data contributor should provide a set of 3D raw image stacks, each of which contains a single neuron or disconnected multiple neurons.
- The image data can be in any popular image format such as 3D TIFF. Before bench-testing, we will convert such images into V3DPBD, a compressed Vaa3D raw file format used in Janelia’s FlyWorkstation and Allen Institute’s In Vitro Single Cell Characterization pipeline, so as to save space and enable consistent measurement of required storage, file I/O time, etc.
- Input image stacks for bench-testing will consist of a single color channel.
- Voxel storage ordering will be first x, then y, then z.
- Raw image data can be in 8-bit or 16-bit, but we will convert all data to 8-bit before bench-testing.
- The output file format for neuron reconstructions will be SWC. The node ordering will be sorted so as to ensure compatibility with popular analysis and visualization tools.
- Each image will be associated with multiple 3D reconstructions linked by a Vaa3D linker file (.ano) file to facilitate visualization and analysis (see "Data Visualization and Analysis Protocol"). Linker files can be edited using any text editor and visualized in Vaa3D by overlaying (or tiling) all associated files in one 3D viewer window.
- A set of 2000 image stacks (download here) are provided as example input image data to port algorithms.
- In case multiple neurons are contained in one image stack, the seed locations for where the tracing should start will be provided as a separate Vaa3D marker file.
- Image data contributed should be accompanied by as much manual annotation as possible, thus to provide a meaningful way to justify the automated reconstructions with respect to the quality of manual annotation.
- Image data contributed to BigNeuron will follow the Open Data agreement (see "Open Data"). Data contributors maintain the ownership of raw image data and will be granted a royalty-free license to use the generated reconstructions for non-profit purposes such as in publications.
- Reconstruction visualization will be supported by Vaa3D. The sorted SWC format reconstruction may also be loaded and visualized in Matlab, CVAPP, neuTube, and many other tools.
- Vaa3D includes a neuron tiling plugin to display many neurons side by side, or on top of each other, thus facilitating comparison of multiple morphologies.
- Vaa3D includes a neuron global feature tool to calculate L-Measure morphological metrics for any reconstruction(s), as well as a “neuron distance” to compute the spatial divergence of two or more neurons.
- The choice of image visualization tools for bench-test data sets will be up to the participating group and include any multi-dimensional data visualizer such as Vaa3D, ImageJ/Fiji, Imaris, etc. Visualization of raw image data sets is not a primary goal of this project and should not be a limiting factor or concern.
Contributed Neuron Image Data Sets
The BigNeuron project has received about 20,000 neuron data sets from a variety of species. We encourage contributions of more data sets of different species and imaging modalities. The following are some representative example data sets received so far:
- Drosophila larval PNS, confocal (B. Ye lab)
- Drosophila adult CNS/brain, confocal (A. Chiang lab)
- Drosophila adult CNS/brain, confocal (G. Jefferis lab)
- Drosophila adult CNS/brain, confocal (G. Rubin lab)
- Dragonfly neurons, confocal (P. Gonzalez-Bellido lab)
- Zebrafish adult retina, confocal (R. Wong lab)
- Zebrafish larval retina, confocal (R. Wong lab)
- Chicken auditory brainstem, confocal (E. Rubel lab)
- Mouse retina, confocal (R. Wong lab)
- Mouse cortical neurons, 2p (J. Kim lab)
- Mouse cortical neurons, 2p (Allen Institute)
- Human cortical neurons, bright field (Allen Institute)
- Human cortical neurons, 2p (Allen Institute)
- Human cortical neurons, bright field (Allen Institute)
Data Contributing Organizations
FlyCircuit.org (Taiwan), Janelia Research Campus, HHMI (USA), Allen Institute for Brain Science (USA), Univ. of Michigan (USA), Univ. of Washington (USA), Univ. of Cambridge (UK), MRC Laboratory of Molecular Biology (UK), Univ. of Florida (USA), Korean Institute of Science and Technology (South Korea), etc.
How to Contribute Neuron Image Data Sets
Please email firstname.lastname@example.org for instructions of data contribution.
The BigNeuron project imposes little restriction on the contribution, use, publication and distribution of the data, while providing as much flexibility as possible for contributors to continue their own work.
BigNeuron calls for the contribution of neuronal images for public domain bench testing. For any contributed neuron images, a set of reconstructions will be produced using the ported automated neuron tracing methods. The data contributors can use such reconstructions freely given the BigNeuron project is appropriately cited.
Contributors of such neuron images of BigNeuron project may use, copy, distribute, publicly perform, publicly display or create derivative works of the neuron reconstructions corresponding to respective images for research, noncommercial, and commercial purposes, given that BigNeuron project as well as the respective reconstructions and algorithms or implementations used in BigNeuron are appropriately referenced. In addition, where the reconstructions contain links to downloadable software applications, services or tools, image contributors may download and use such applications, services or tools as long as such contributors adhere to any license terms and conditions provided with those applications, services or tools.
Image contributors may not post reconstructions on social media or other third-party websites that require image contributors to acknowledge that they own the data they post (e.g., YouTube, Flickr and Twitter). Image contributors agree that they will not use the BigNeuron data in any manner that would violate anyone else's rights, such as copyright, trademark, patent, privacy or other rights. This includes removing any copyright, trademark or other proprietary notices from the reconstructions. Image contributors may not create hyperlinks to the BigNeuron resources (e.g. website, database, documentation site) that portray this project in a false or misleading light. Image contributors agree that they will only make lawful use of the BigNeuron Data and Resources in compliance with all federal, state, and local laws and regulations.
Image contributors may, and are encouraged to, develop new methods, applications, interfaces or other inventions or works that improve the use of, and build upon, the reconstructions. In order to make the reconstructions available to image contributors and others, however, the BigNeuron project must preserve its freedom to innovate. If image contributors develop improvements based on or utilizing the reconstructions, and such image contributors obtain any proprietary rights in or to that improvement, the image contributors and their successors or assigns agree that image contributors will not assert any claim for infringement against the BigNeuron project for the use of any improvement that was independently developed by or on behalf of the BigNeuron project. Additionally, the BigNeuron project retains its rights, title and interest in any reconstructions that are part of or are used by image contributors to create an Improvement.
The “BigNeuron Terms of Participation and Use” should be applied to image contributors.