Supplementary Figure 3: PyKNOSSOS.
From: Dense EM-based reconstruction of the interglomerular projectome in the zebrafish olfactory bulb
Screenshot of PyKNOSSOS in tracing mode. The four viewports on the left show reslices through the original image stack (here, the upper left shows the x-y plane, the lower right is orthogonal to the traced process shown in red, and the remaining two viewports are orthogonal to the lower right viewport. Other viewport assignments are possible). The fifths viewport shows reconstructions. Zoom levels, translations, rotations, and superimpositions with image data can be adjusted in real time. Panels on the right provide access to different functions for tracing, annotation and visualization (see manual in Supplementary Methods). A tab for synapse annotation is shown on the lower right. Images in the viewports are from the olfactory bulb of a zebrafish larva (4.5 dpf) and were acquired with a vpSEM in high vacuum (EL = 2 keV, De = 17.5 e–nm–2, 9.25 × 9.25 × 25 nm voxels).
Ergonomic software is important to minimize the amount of human labor in circuit reconstruction. We generated a software package for manual skeleton tracing, synapse annotation and visualization with two goals in mind. First, it should be written in a widely used programming language to facilitate modifications and extensions by users. Second, it should allow for the real-time visualization of large and complex reconstructions (hundreds of neurons) together with raw image data. Existing software packages used for neuron reconstruction include KNOSSOS (www.knossostool.org; Helmstaedter, M., et al. [2011] Nat. Neurosci. 14, 1081-1088), CATMAID (http://catmaid.org/; Saalfeld, S., et al. [2009] Bioinformatics 25, 1984-1986), ilastik (www.ilastik.org; Sommer, C., et al. [2011] 8th Internat. Symp. Biomed. Imaging [ISBI] Proceedings, 230-233), and others. KNOSSOS and CATMAID are specifically designed for large-scale reconstruction projects. Inspired by these tools, we created new software that is written in PYTHON and uses the VTK library for 3D visualization (http://www.vtk.org/). The software was developed by close interactions between programmer and user and is named PyKNOSSOS because the user interface is closely related to KNOSSOS. A manual with screenshots can be found in Supplementary Software Information. The code for PyKNOSSOS is available at https://github.com/adwanner/PyKNOSSOS. Further information is available at www.ariadne-service.ch.
Important features of PyKNOSSOS include the following:
1. Efficient rendering of reconstructed neurons using VTK. Using the visualization toolkit (VTK; http://www.vtk.org/), PyKNOSSOS can interactively visualize hundreds of reconstructed neurons in 3D on a standard desktop computer together with the underlying image data. This functionality is important for efficient data browsing, tracing and error correction. For higher-level analyses, visualization in PyKNOSSOS can be integrated into custom workflows. Moreover, PyKNOSSOS can be used to generate 3D renderings and animated displays of raw data and reconstructions.
2. Multi-resolution view. Like KNOSSOS, PyKNOSSOS dynamically loads cubes of data into RAM as users navigate through a volume. This allows users to browse through large datasets (>1TB) with minimal RAM requirements. Storing multiple sets of cubes at different resolutions allows for seamless zooming through large volumes and the extraction of virtual reslices with arbitrary orientation and zoom-level.
3. Virtual reslicing of the raw data orthogonal to local processes. Tracing of neurites and branch point detection should be particularly efficient when users view a section through the EM data volume that is orthogonal to the process being traced. Such virtual reslices should also facilitate the identification of synapses because presynaptic vesicles, the synaptic density, and the postsynapse are contained in the same view. PyKNOSSOS automatically calculates a rotation-minimized “locally orthogonal” section (Wang, W., et al. [2008] ACM Trans. Graph. 27, 2) during tracing and presents it in a separate viewport, in addition to the cardinal cross-sections and the imaging plane. We found that the “locally orthogonal” view facilitates the detection of branch points and increases tracing speed.
4. Synapse annotation tools. PyKNOSSOS includes tools to define the location and direction of a synapse by three successive clicks on the presynaptic process, the synaptic density, and the postsynaptic process. In addition, synapses can be assigned to user-defined classes by a confidence level. Furthermore, pre-calculated “flight paths” can be loaded to automatically visit all branches of a reconstructed neuron. Using this mode, only two clicks on the synaptic density and postsynapse are required to define a synaptic connection. These features, together with the “locally orthogonal” view, facilitate manual synapse annotation. Optionally, each synapse can be assigned to a predefined class and annotated with a confidence level.
Note that some of these functionalities have recently been incorporated also into other software packages.
