HPC Leading Edge News
HPC Leading Edge Resources
November 9, 2018
The 2017 Nobel Prize in Chemistry was awarded to Jacques Dubochet, Joachim Frank and Richard Henderson for the development of cryo-electron microscopy (cryo-EM), which both simplifies and improves the imaging of biomolecules.
In awarding the prize, the Royal Swedish Academy of Sciences said cryo-EM has moved biochemistry into a new era. “Researchers can now freeze biomolecules mid-movement and visualize processes they have never previously seen, which is decisive for both the basic understanding of life’s chemistry and for the development of pharmaceuticals,” the Academy said.
However, the new technology is not without its challenges when implementing it to analyze biomolecular structures. “Because biomolecules are highly sensitive to radiation damage by the electron beam, the molecular images have to be taken at a low dose that gives rise to an extremely high degree of noise in the formation of the image,” writes Youdong (Jack) Mao, an instructor in the Department of Cancer Immunology and AIDS, Dana-Farber Cancer Institute, Department of Microbiology and Immunobiology, Harvard Medical School. “This situation leads to one of the most critical challenges facing computational approaches to cryo-EM reconstruction of biomolecules; namely, the extraction of signal from heavy noise.”
HPC Custs through the Noise
Mao’s work requires a large number of very noisy images to be analyzed to reconstruct the structure of a molecule up to atomic resolution via averaging and statistical techniques. This is a highly data-intensive and computationally demanding process.
“The HPC clusters from Dell EMC are critical to our research missions that highly depend on the analysis of big data generated from highly automated cryo-electron microscopes,” says Mao, who is also director of the Intel® Parallel Computing Center for Structural Biology (PCCSB). The Intel PCCSB is home to the Sullivan supercomputer, which is composed of a high-performance RAID storage sub-system of a hundred terabytes, a sub-system powered by thousands of Intel® Xeon Processor cores, and a sub-system powered by the latest Intel® Xeon Phi™ coprocessors that incorporate several thousand X86 CPU cores. “The HPC systems facilitate the development of state-of-the-art algorithms in pursuit of structural solutions to those grand biomedical problems, which would deliver innovations in cancer immunotherapy and precision medicine.”
The Road to ROME
Many efforts at the Intel PCCSB revolve around refining a software-hardware system that uses machine learning for massively parallel cryo-EM data processing. The ROME (Refinement and Optimization via Machine lEarning for cryo-EM) software package is the result of those efforts.
Optimized on both Intel Xeon multi-core CPUs and Intel Xeon Phi many-core coprocessors, version 1.0 of the ROME system continues to be improved. Upgrades are planned to introduce more advanced machine-learning algorithms for high-performance cryo-EM data analysis, including nonlinear dimensionality reduction, deep learning and 4D reconstructions, according to Intel PCCSB.