Anne E. Carpenter - US grants

DESCRIPTION (provided by applicant): Most laboratories studying biological processes and human disease use light/fluorescence microscopes to image cells and other biological samples. There is strong and growing demand for software to analyze these images, as automated microscopes collect images faster than can be examined by eye and the information sought from images is increasingly quantitative and complex. We have begun to address this demand with CellProfiler, a versatile, open-source software tool for quantifying data from biological images, particularly in high-throughput experiments (www.cellprofiler.org). CellProfiler can extract valuable biological information from images quickly while increasing the objectivity and statistical power of assays. In the three years since its release, it has become widely used, having been downloaded more than 8,000 times by users in over 60 countries. Using CellProfiler's point-and-click interface, researchers build a customized chain of image analysis modules to identify and measure biological objects in images. The software evolved in an intensely collaborative and interdisciplinary research environment with dozens of ongoing projects and has been successfully applied to a wide range of biological samples and assays, from counting cells to scoring complex phenotypes by machine learning. To enable further biological imaging research, meet the needs of the growing user base, and expand the community that benefits from CellProfiler, we propose to improve its capabilities, interface, and support: First, we will add user-requested capabilities, leveraging existing open-source projects by interoperating with them where feasible. These new features will include object tracking in time-lapse movies, compatibility with additional file formats, new image processing algorithms, and expanded tools for quality control, performance evaluation, cluster computing, and workflow management. Second, we will improve the interface, increase processing speed, and simplify the addition of new features by refactoring and porting the MATLAB-based code to an open-source language and instituting proven software development practices. Third, we will provide user, educator, and developer support and outreach for CellProfiler. These activities will facilitate research in the scientific community and help guide usability improvements. These improvements to the only open-source software for modular, high-throughput biological image analysis will enable hundreds of NIH-funded laboratories to make high-impact biological discoveries from cell images across all disciplines within biology. PUBLIC HEALTH RELEVANCE:

DESCRIPTION (provided by applicant): High-throughput screening (HTS) is a technique for searching large libraries of chemical or genetic perturbants, to find new treatments for a disease or to better understand disease pathways. As automated image analysis for cultured cells has improved, microscopy has emerged as one of the most powerful and informative ways to analyze screening samples. However, many diseases and biological pathways can be better studied in whole animals-particularly diseases that involve organ systems and multicellular interactions, such as metabolism and infection. The worm Caenorhabditis elegans is a well-established and effective model organism, used by thousands of researchers worldwide to study complex biological processes. Samples of C. elegans can be robotically prepared and imaged by high-throughput microscopy, but existing image-analysis methods are insuf- ficient for most assays. In this project, image-analysis algorithms that are capable of scoring high-throughput assays of C. elegans will be developed. The algorithms will be tested and refined in three high-throughput screens, which will uncover chemical and genetic regulators of fat metabolism and infection: (1) A C. elegans viability assay to identify modulators of infection. The proposed algorithms use a probabilistic shape model of C. elegans in order to identify and mea- sure individual worms even when the animals touch or cross. These methods are the basis for quantifying many other phenotypes, including body morphology and subtle variations in reporter signal levels. (2) A C. elegans lipid assay to identify genes that regulate fat metabolism. The algorithms proposed for illumination correction, level-set-based foreground segmentation, well-edge detection, and artifact removal will result in improved or- business in high-throughput experiments. (3) A fluorescence gene expression assay to identify regulators of the response of the C. elegans host to Staphylococcus aureus infection. The proposed techniques for constructing anatomical maps of C. elegans will make it possible to quantify a variety of changes in fluorescent localization patterns in a biologically relevant way. In addition to discovering new metabolism- and infection-related drugs and genetic regulators through these specific screens, this work will provide the C. elegans community with (a) a new framework for extracting mor- phological features from C. elegans for quantitative analysis of this organism, and (b) a versatile, modular, open-source toolbox of algorithms enabling the discovery of genetic pathways, chemical probes, and drug can- didates in whole organism high-throughput screens relevant to a variety of diseases. This work is a close collaboration with C. elegans experts Fred Ausubel and Gary Ruvkun at Massachusetts General Hospital/Harvard Medical School, with Polina Golland and Tammy Riklin-Raviv, experts in model-based segmentation and statistical image analysis at MIT's Computer Science and Artificial Intelligence Laboratory, and with Anne Carpenter, developer of open-source image analysis software at the Broad Institute.

Project Summary Most laboratories studying biological processes and human disease use microscopes to image samples. Whether in small­ or large­scale microscopy experiments, biologists increasingly need software to identify and measure cells and other biological entities in images, to improve speed, objectivity, and/or statistical power. The principal investigator envisions bringing transformative image analysis and machine learning algorithms and software to a wide swath of biomedical researchers. In a decade, researchers will tackle fundamentally new problems with quantitative image analysis, using seamless imaging workflows that have dramatic new capabilities going beyond the constraints of human vision. To this end, the PI will collaborate with biologists on important quantitative imaging projects that also yield major advancements to their open­source image analysis software, CellProfiler. This versatile, user­friendly software is indispensable for biomedical research. Launched 125,000+ times/year worldwide, it is cited in 3,400+ papers from 1,000+ laboratories, impacting a huge variety of biomedical fields via assays from counting cells to scoring complex phenotypes by machine learning. CellProfiler evolves in an intensely collaborative and interdisciplinary research environment that has yielded dozens of discoveries and several potential drugs. Still, many biologists are missing out on the quantitative bioimaging revolution due to lack of effective algorithms and usable software for their needs. In addition to maintaining and supporting CellProfiler, the team will implement biologist­requested features, algorithms, and interoperability to cope with the changing land­ scape of microscopy experiments. Challenges include increases in scale (sometimes millions of images), size (20+ GB images), and dimensionality (time­lapse, three­dimensional, multi­spectral). Researchers also need to accommodate a variety of modalities (super­resolution, single­molecule, and others) and integrate image analysis into complex workflows with other software for microscope control, cloud computing, and data mining. The PI will also pioneer novel algorithms and approaches changing the way images are used in biology, including: (1) a fundamental redesign of the image processing workflow for biologists, leveraging revolutionary advancements in deep learning, (2) image analysis for more physiologically relevant systems, such as model organisms, human tissue samples, and patient­derived cultures, and (3) data visualization and interpretation software for high­dimensional single­cell morphological profiling. In profiling, subtle patterns of morphological changes in cells are detected to identify causes and treatments for various diseases. We will also (4) integrate multiple profiling data types: morphology with gene expression, epigenetics, and proteomics. Ultimately, we aim to make perturbations in cell morphology as computable as other large­scale functional genomics data. Overall, the laboratory?s research will yield high­impact discoveries from microscopy images, and its software will enable hundreds of other NIH­funded laboratories to do the same, across all biological disciplines.

Project Summary    Most laboratories studying biological processes and human disease use microscopes to image samples.  Whether in small­ or large­scale microscopy experiments, biologists increasingly need software to identify and  measure cells and other biological entities in images, to improve speed, objectivity, and/or statistical power.   The principal investigator envisions bringing transformative image analysis and machine learning algorithms  and software to a wide swath of biomedical researchers. In a decade, researchers will tackle fundamentally  new problems with quantitative image analysis, using seamless imaging workflows that have dramatic new  capabilities going beyond the constraints of human vision.  To this end, the PI will collaborate with biologists on important quantitative imaging projects that also yield  major advancements to their open­source image analysis software, CellProfiler. This versatile, user­friendly  software is indispensable for biomedical research. Launched 125,000+ times/year worldwide, it is cited in  3,400+ papers from 1,000+ laboratories, impacting a huge variety of biomedical fields via assays from counting  cells to scoring complex phenotypes by machine learning. CellProfiler evolves in an intensely collaborative and  interdisciplinary research environment that has yielded dozens of discoveries and several potential drugs.  Still, many biologists are missing out on the quantitative bioimaging revolution due to lack of effective  algorithms and usable software for their needs. In addition to maintaining and supporting CellProfiler, the team  will implement biologist­requested features, algorithms, and interoperability to cope with the changing land­  scape of microscopy experiments. Challenges include increases in scale (sometimes millions of images), size  (20+ GB images), and dimensionality (time­lapse, three­dimensional, multi­spectral). Researchers also need to  accommodate a variety of modalities (super­resolution, single­molecule, and others) and integrate image  analysis into complex workflows with other software for microscope control, cloud computing, and data mining.   The PI will also pioneer novel algorithms and approaches changing the way images are used in biology,  including: (1) a fundamental redesign of the image processing workflow for biologists, leveraging revolutionary  advancements in deep learning, (2) image analysis for more physiologically relevant systems, such as model  organisms, human tissue samples, and patient­derived cultures, and (3) data visualization and interpretation  software for high­dimensional single­cell morphological profiling. In profiling, subtle patterns of morphological  changes in cells are detected to identify causes and treatments for various diseases. We will also (4) integrate  multiple profiling data types: morphology with gene expression, epigenetics, and proteomics. Ultimately, we  aim to make perturbations in cell morphology as computable as other large­scale functional genomics data.  Overall, the laboratory?s research will yield high­impact discoveries from microscopy images, and its  software will enable hundreds of other NIH­funded laboratories to do the same, across all biological disciplines.        

Project Summary The Center for Open Bioimage Analysis will serve the cell biology community?s growing need for sophisticated software for light microscopy image analysis. Quantitative image analysis has become an indispensable tool for biologists using microscopy throughout basic biological and biomedical research. Quantifying images is now a critical, widespread need as imaging experiments continue to grow in scale, size, dimensionality, scope, modality, and complexity. Many biologists are missing out on the quantitative bioimaging revolution due to lack of effective algorithms and/or usable software for their needs, or lack of access to training. The Center brings together the Carpenter laboratory at the Broad Institute and the Eliceiri laboratory at the University of Wisconsin­Madison, and in doing so brings together the two most popular open source bioimage analysis projects, ImageJ (including ImageJ2 and FIJI) and CellProfiler. Through the collaborative development and dissemination of open source image analysis software, as well as training events and resources, the Center will empower thousands of researchers to apply advanced analytics in innovative ways to address new experimental areas. Building on the team?s expertise developing algorithms and user­friendly software for use in biology under real­world conditions, the Center will focus on two Technology Research and Development (TR&D) projects: deep learning­based image processing, and accessibility of image­processing algorithms for biologists. This work will not occur in isolation at the Center? rather, the Center will nucleate a larger community working on these two areas and serve as a catalyst and organizing force to create software and resources shared by all. The Driving Biological Projects (DBPs) will serve a major role in driving the TR&D work: our teams are accustomed to working deeply and iteratively on problems side by side and with frequent feedback from biologists. This will ensure that important cell biological problems drive the work of the Center. The DBPs reflect tremendous variety in terms of biological questions, model systems, imaging modalities, and researcher expertise and will ensure robustness of our tools for the widest possible impact on the community. Continuing the teams? track record with ImageJ and CellProfiler, two mature open source bioimage analysis software projects critical to the work of biologists worldwide, the Center will also assist and train biologists in applying the latest computational techniques to important biological problems involving images. In short, the need for robust, accurate, and readily usable software is more urgent than ever. The Center for Open Bioimage Analysis will serve as a hub for pioneering new computational strategies for diverse biological problems, translating them into user­friendly software, further developing ImageJ and CellProfiler, and training the biological community to apply advanced software to important and diverse problems in cell biology.