Integrating the latest breakthroughs in biochemistry, high performance computing, optical processing, and storage is enabling remarkable advances in the fields of healthcare, drug discovery, and genomic research. Together, they are creating exciting personal therapeutic strategies for living longer, healthier lifestyles that were unimaginable until now.
Genomics technology is now capable of determining a person’s predisposition to threatening diseases like cancer, heart disease, diabetes and more. Personal healthcare information that was previously unattainable or unaffordable can be purchased for the price of picking up the tab for a fine dinner with friends.
The recent surge in consumer genomics has led to knowledge about oneself that is staggering, but it’s just the beginning. Imagine having your entire genome information at your fingertips, helping you and your doctor make informed, customized decisions about drug dosages and their effectiveness in preventing and curing diseases for your body. For instance, a man could learn whether taking regular doses of Finasteride, would reduce or prevent the possibility of prostate cancer, while feeling confident that the drug would cause few or no side effects. But this is only the beginning of the many life science achievements forecast to take place within the next few years, such as new designer drugs genetically tailored for specific genotypes.
High performance computing technologies will be at the forefront of the personal healthcare revolution, making it possible to realize and accelerate radical medical breakthroughs.
New Technology Collapsing Costs
To reach this objective, next- and “3rd” generation gene sequencers will be key technology components that exponentially drive down the costs of determining one’s entire personal genome. Costs are plummeting because the process for creating genome data is being radically affected by the newest technologies incorporating high performance compute capabilities in the workflow.
Without next/third generation gene sequencers coupled to accelerated computing technologies and parallel processing algorithms, such advances in the “personal” healthcare industry might not be possible, at least not in a reasonable time to maximize the sequencing analysis to affect peoples’ lives in the near future.
With HPC capability comes the hyper-exponential growth of digital information in the form of useful yet extensive and unruly datasets. Beyond capture and create, the secondary analysis of this astronomical data growth will push even the newest computing technologies to their physical and practical limits. Fortunately, as sequencing instruments break new ground, so do the compute acceleration technologies designed to tackle these new bioinformatics challenges.
Choosing the Appropriate HPC Platform
The nearly limitless range of bioscience research application requirements for various accelerated hybrid-computing technologies leaves room for choice and error when considering appropriate systems. For many researchers and technologists, choosing the best HPC and acceleration technologies can be daunting because they have not yet been fully implemented in standard, commercially-available mainstream platforms. Consequently, bioinformaticists, drug developers and IT specialists have to be close in mindset and cross-functional in their disciplines to determine and optimize the correct mix of hardware and software.
However, it is ill advised to wait for these systems to become commodities since cutting-edge, competitive research and results will benefit from adopting acceleration solutions now. Some risk is often involved in early technology adoption, but not getting past the starting points for evaluation will potentially create a greater risk of loosing a competitive advantage, a risk far outweighing immediate concerns about which accelerated computing solution to select.
Paths to Success
A few distinctively differing acceleration technologies may currently be applied, including:
Field-programmable gate array (FPGA)-accelerated hybrid computing systems based on in-socket accelerators installed in standard, low-profile or blade servers, general-purpose graphics processing unit (GPGPU) as plug-in PCIe bus accelerators in standard servers and workstations, and accelerators like the Cell processor, the brains of the Sony PlayStation3 and among the most notable HPC implementations for the fastest supercomputer, Roadrunner.
The FPGA In-Socket Solution
Historically, FPGA-based acceleration systems have a good track record for accelerating the most challenging bioinformatics codes, like sequence search and detection, sequence comparison, mapping and alignment. This use has been steadily growing because both FPGA densities and their internal bandwidth are making them exceptional choices for accelerating fine-grained, variable-bit width operations, including binary, fixed point and integer calculations, which are perfectly matched for genomics problems.
The proof points of these types of codes have achieved orders of magnitude performance increases versus comparable, un-accelerated systems. Specifically codes like Smith Waterman based SSEARCH, most derivatives of Clustal and BLAST alignment codes, Hidden Markov Model based HMMER, and FASTA codes for protein alignment can be easily accelerated with excellent results. In many instances, a minimum of 10X performance gains are achieved and 30X-60X are possible depending on the algorithms used.
As scientific developments rapidly change, so do the needs for development platforms based on standard and flexible hardware and software programming combinations, or true hybrid computers. In the past, FPGA acceleration has been limited to highly-proprietary hardware with HDL programming, a key deficiency in meeting the fast-changing bioinformatic algorithm developers’ requirement.
Now there are affordable, reliable HPC platforms with much easier to use parallel programming languages and specific virtual processors, allowing the developer to focus on his application instead of reprogramming and designing the hardware circuits in FPGAs. This is a game changer in FPGA acceleration utility.
To complete the hybrid computer, there are in-socket co-processor accelerators that connect FPGAs to the main CPU with supported standard high-throughput interfaces. Coupled with effective parallel software programming that transcends this accelerator choice from a black box hardware-centric design to an open software-programming environment. Acceleration techniques can soon be realized on new extensions of the most popular codes mentioned plus Eland, Maq, SOAP, Mosaik, and many other proprietary and novel next generation sequence (NGS) codes to analyze DNA and proteins.
FPGA accelerated hybrid computing systems are a good choice for these text-matching codes and their associated growing datasets, particularly when total cost of operation and low power are requirements for a greener work space.
Floating Point Solutions with GPGPU and Cell Co-Processors
In life science applications, general purpose graphics processing units (GPGPUs) and Cell co-processors have proven themselves effective in analyzing molecular mechanical codes and in image analysis. These co-processors are attractive options for accelerating floating-point-intensive applications resulting in orders of magnitude performance increases for mathematical computations. The rising adoption is occurring because GPGPUs and Cell co-processors are now capable of performing in applications beyond their original design purposes in graphics and gaming.
GPUs and Cell take different approaches to performance gains. GPUs accomplish this with a manycore architecture, while Cell is a heterogeneous, mixed multicore architecture. Unlike FPGAs, they are created with many hard-wired floating point (FP) units, making them ideal for accelerating most single-precision and some double-precision numerical algorithms.
In life sciences, there are many molecular mechanical codes used for visualizing molecular docking and solving atom-to-atom interactions for drug discovery. GPGPUs have shown proof to accelerated codes like NAMD (NAnoscale Molecular Dynamics), VMD (Visual Molecular Dynamics), and CHARMM (Chemistry at Harvard Macromolecular Mechanics). Cell has been proven to effectively accelerate codes like GROMACS (GROningen Machine for Chemical Simulations), a protein folding code that unlocks the mystery of protein assembly and its relationship to cancers, Parkinson’s disease and Alzheimer’s. These, and similar codes benefit from a hybrid mix of CPU and floating-point acceleration.
Other relevant life science applications for floating point acceleration are image processing and analysis. Codes that convert genome sequencing reagent reactions in the form of camera images perform image filtering, sharpening and enhancing in real time.
Math-intensive codes make PCIe-based GPGPUs a practical hardware solution choice since simply plugging them into expensive sequencing instruments accelerates results with little change to the hardware. These instruments can absorb the additional power required and acoustic envelope tradeoffs for performance increases.
The Cell processor is also applicable for accelerating medical imaging codes used for CT Scan, X-Ray, ultrasound and MRI workflows. Cell’s additional advantage of being a low power solution shows promise for advancing technologies for smaller, image-guided surgery engines. It’s also proven to scale across large power-efficient supercomputers for advanced simulations and visualization.
Choosing a Hybrid Computer System Today
In the past, choosing the appropriate HPC solution for life scientists was difficult because the parallel programming model, the CPU interface logic, and the FPGA in-socket solutions were not fully developed. Today you can acquire simple hybrid computing servers and start developing. Now, parallel programming solutions, FPGA virtual algorithm processors, and working operating systems are available for both FPGA and floating-point accelerators.
Selecting the optimum hybrid computing solution for a given life science application might seem confusing at first since the various acceleration technologies can perform well for a wide range of scientific computing. Given a diligent coding effort, floating-point accelerators can tackle pattern-matching codes and FPGAs can do floating point and imaging. The key to selecting the right HPC solution is to understand where each respective technology has proven early success, as this provides strong clues about their respective platform strengths and weaknesses.
Into the Future
Building these hybrid architectures will be an absolute requirement for meeting the exponentially-increasing R&D demands in life sciences. For instance, massive hybrid supercomputer clusters will likely redefine the world’s fastest genome computers. With the help of innovative technology partners, the largest systems vendors are working on this now.
Soon “multi-hybrid” computers will incorporate multiple, complementary acceleration types working together in a single chassis or many clusters including an OS, language and API support. It will be possible to incorporate multiple, differing, accelerated workflows, each having a dedicated or reconfigurable acceleration type optimized for each application. This is not at all far-fetched because a multiple, heterogeneous mix of processing types, all working in unison, is a model easily recognized in many of today’s computerized systems, such as network switches, PDAs, auto electronics and storage products.
When this multi-hybrid systems technology becomes commonly available, choosing an accelerator type wont feel like a life or death decision…or will it?
Author the Author
Mike Calise is the executive vice president and US general manager at Mitrionics and has 25 years experience in the semiconductor and systems industry. Prior to Mitrionics, he was president of ClearSpeed and has held various sales, marketing and senior management positions for companies including Intel Corporation, Benchmarq Microelectronics (Texas Instruments), Catalyst Semiconductor, start-up SOC IP providers, Palmchip and Improv Systems. He received his B.S.E.E. from University of Buffalo in 1983.