Scientific Program Manager Allen Institute Seattle, Washington, United States
Abstract: The Allen Institute for Cell Science aims to elucidate how human induced pluripotent stem cells (hiPSCs) establish and maintain dynamic cellular structures and transition between states during differentiation and disease. A crucial step towards this goal is being able to quantify changes accurately and reproducibly in the size and shape of the cell and subcellular organelles. We took advantage of high-resolution 3D microscopy images of fluorescently tagged human induced pluripotent stem cells (hiPSCs) from the Allen Cell Collection to develop a suite of image-analysis tools. This includes the Allen Cell and Structure Segmenter, a Python-based open-source toolkit for 3D segmentation of fluorescence microscope images, which we have updated and made more accessible by developing napari plugins for each the Classic Image Segmentation workflow and the Iterative Deep Learning Workflow. The latter is powered by a comprehensive deep learning tool for streamlining microscopy image transformations called CytoDL which allows integration of multiple workflows. Already integrated is MegaSeg, a 3D convolutional neural network segmentation model trained on a large dataset of high-quality 3D fluorescent images and corresponding segmentations for 13 intracellular structures with distinct morphologies. MegaSeg enables robust, morphology-agnostic deep feature extraction that generalizes across various intracellular structures and permits their accurate 3D segmentations. To visualize these data, we recently released the Timelapse Feature Explorer, a web-based application designed for the interactive visualization and analysis of segmented time-series microscopy data. Finally, to address the challenges associated with handling these large datasets, we developed BioFile Finder, an advanced, web-based application designed to make data easy to search, aggregate, and share in ways that promote novel insights and reproducible science while reducing redundant pixel storage. This open-source tool provides a robust search engine that allows users to quickly locate specific images from user-supplied or publicly available datasets, accelerating data analysis, management, and discovery. These tools, resources, and data and more can be found at allencell.org.
