Accelerating Discovery in 3D Microanalysis: Leveraging Open Source Software and Deskside High Performance Computing. Yoo, T., S., Lowekamp, B., C., Kuybeda, O., Narayan, K., Frank, G., A., Bartesaghi, A., Borgnia, M., Subramaniam, S., Sapiro, G., & Ackerman, M., J. Microsc. Microanal, 2018.
Accelerating Discovery in 3D Microanalysis: Leveraging Open Source Software and Deskside High Performance Computing [link]Website  abstract   bibtex   
The recent decade has seen a dramatic elevation in the computing power affordably and routinely available to biological laboratories. Computing cores in servers, desk-side, and even laptop computers have doubled in number and capability on the order of every two years, making workstations today the rivals of supercomputers from the year 2000. 64-bit processors have substantially increased addressable main memory, and computational analysis can now keep pace with the growing size of datasets. Commensurately, the operation of the microscopes has become increasing sophisticated, evolving from analog consoles to digital interfaces. These developments enable the automation of image collection, storage and quantitative analysis of the resulting data. Inexpensive storage and high bandwidth in digital networks promote the sharing of data and broad multidisciplinary interaction among research groups. We present examples of how our collaboration between high performance computing developers at the National Library of Medicine and high-resolution electron microscopists at the National Cancer Institute are working to advance and develop new science. Our system designers combine innovative software and commercial computing hardware such as graphics processing units (GPUs) originally developed for entertainment to accelerate the solving of complex problems in microbiology. We concentrate much of our effort on the use of open source software, developed in part by our team or sponsored by our programs and repurposed for microanalysis [1]. Super-resolution through sub-volume averaging:In recent years, our group has been developing methods in transmission electron tomography to resolve glycoprotein complexes on the surfaces of viruses at the scale of a single nanometer. By accurately aligning and averaging very noisy sub-volumes we decode the structure of molecules at the edge of the resolution of our instruments. Starting with the early successes of our team, we have redesigned our software to use a combination of conventional code (written in MATLAB using the Parallel Toolbox) and highly parallel subroutines designed as NVIDIA CUDA kernels for singular value decomposition, programmed in C++, and run on multiple GPUs. The resulting architecture can generate results an order of magnitude faster than previously achieved using hardware that costs less than a tenth the price of earlier computing clusters. Results that previously took days can be achieved in hours using machines that comfortably fit in a laboratory budget [2]. Segmentation and visualization using open source software:Software originally designed for computer aided diagnosis and computer assisted surgery can be profitably applied to 3D microanalysis. We have adapted algorithms and filters from the open source Insight Toolkit (ITK) for the study of data from FIB-SEM data. Further, we have adapted 3DSlicer [3], a free open source medical image analysis console for quantitative research to 3D high-resolution microanalysis. We supplement these efforts by
@article{
 title = {Accelerating Discovery in 3D Microanalysis: Leveraging Open Source Software and Deskside High Performance Computing},
 type = {article},
 year = {2018},
 volume = {20},
 websites = {https://www.cambridge.org/core/services/aop-cambridge-core/content/view/S1431927614005595},
 id = {c7f27b2e-23d4-3f1d-825c-5d78f584f634},
 created = {2018-01-19T16:05:38.730Z},
 accessed = {2018-01-19},
 file_attached = {false},
 profile_id = {5477662f-c7a8-37b0-902d-765da20fee71},
 last_modified = {2018-03-08T04:36:31.566Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {644af951-328a-4794-8436-dbab34dfd4ee},
 private_publication = {false},
 abstract = {The recent decade has seen a dramatic elevation in the computing power affordably and routinely available to biological laboratories. Computing cores in servers, desk-side, and even laptop computers have doubled in number and capability on the order of every two years, making workstations today the rivals of supercomputers from the year 2000. 64-bit processors have substantially increased addressable main memory, and computational analysis can now keep pace with the growing size of datasets. Commensurately, the operation of the microscopes has become increasing sophisticated, evolving from analog consoles to digital interfaces. These developments enable the automation of image collection, storage and quantitative analysis of the resulting data. Inexpensive storage and high bandwidth in digital networks promote the sharing of data and broad multidisciplinary interaction among research groups. We present examples of how our collaboration between high performance computing developers at the National Library of Medicine and high-resolution electron microscopists at the National Cancer Institute are working to advance and develop new science. Our system designers combine innovative software and commercial computing hardware such as graphics processing units (GPUs) originally developed for entertainment to accelerate the solving of complex problems in microbiology. We concentrate much of our effort on the use of open source software, developed in part by our team or sponsored by our programs and repurposed for microanalysis [1]. Super-resolution through sub-volume averaging:In recent years, our group has been developing methods in transmission electron tomography to resolve glycoprotein complexes on the surfaces of viruses at the scale of a single nanometer. By accurately aligning and averaging very noisy sub-volumes we decode the structure of molecules at the edge of the resolution of our instruments. Starting with the early successes of our team, we have redesigned our software to use a combination of conventional code (written in MATLAB using the Parallel Toolbox) and highly parallel subroutines designed as NVIDIA CUDA kernels for singular value decomposition, programmed in C++, and run on multiple GPUs. The resulting architecture can generate results an order of magnitude faster than previously achieved using hardware that costs less than a tenth the price of earlier computing clusters. Results that previously took days can be achieved in hours using machines that comfortably fit in a laboratory budget [2]. Segmentation and visualization using open source software:Software originally designed for computer aided diagnosis and computer assisted surgery can be profitably applied to 3D microanalysis. We have adapted algorithms and filters from the open source Insight Toolkit (ITK) for the study of data from FIB-SEM data. Further, we have adapted 3DSlicer [3], a free open source medical image analysis console for quantitative research to 3D high-resolution microanalysis. We supplement these efforts by},
 bibtype = {article},
 author = {Yoo, Terry S and Lowekamp, Bradley C and Kuybeda, Oleg and Narayan, Kedar and Frank, Gabriel A and Bartesaghi, Alberto and Borgnia, Mario and Subramaniam, Sriram and Sapiro, Guillermo and Ackerman, Michael J},
 journal = {Microsc. Microanal}
}
Downloads: 0