SDSC, TimeLogic and Sun validate ultra-fast Hidden Markov model analysisCrystal Bay 08 May 2002 Sun Microsystems, TimeLogic Corporation, manufacturer of the DeCypher line of biocomputing accelerators, and the San Diego Supercomputer Center (SDSC) at the University of California, San Diego (UCSD) recently completed benchmark testing and scientific validation of large-scale Hidden Markov Model bioinformatics analyses comparing speed and results to those obtained on SDSC's Linux PC cluster.
In the tests, designed by Adam Godzik, Ph.D., a researcher at SDSC in the Joint Center for Structural Genomics (JCSG) and associate professor and programme director for Bioinformatics and Biological Complexity at The Burnham Institute, compared a cluster of 32 x 1-GHz Pentium III Linux PCs running Dr. Sean Eddy's (Washington University at St. Louis) popular HMMER software, to an 8 CPU Sun Fire 6800 server containing the DeCypher XD-4G FPGA reconfigurable computing processor array on the Sun Fire's PCI bus.
The benchmark employed NCBI's NR75 non-redundant protein database containing 372,119 sequences (128.1 million aa symbols) as the query set, and SDSC's HMM database of 19,192 models as the target of the search. Fully utilising the Linux cluster, this analysis requires 144 days run time, approximately five months of dedicated, uninterrupted processing.
The DeCypher accelerated Sun Fire machine completed the entire task in just 41 hours 46 minutes. Validity of scientific results was confirmed by SDSC scientists, based on comparison to those of Dr. Eddy's software. SDSC researcher Slawek K. Grzechnik, technical coordinator for the JCSG Bioinformatics Core, conducted the testing.
Commenting on the dramatic result, Jim Lindelien, founder and CTO of TimeLogic, said: "This benchmark ran 82-fold faster than the Linux cluster's 32 CPUs. That works out to 2,624 CPUs equivalent throughput for this important bioinformatics application, in the footprint and power demand of a single Sun machine, with impressive price-performance improvement over what is widely thought to be the most cost-effective technology available."
The Joint Center for Structural Genomics (JCSG), a consortium of California scientific research organisations including The Scripps Research Institute (TSRI)/Genomics Institute for the Novartis Research Foundation (GNF); SDSC; and the Stanford Synchrotron Radiation Laboratory (SSRL, a Division of the Stanford Linear Accelerator Center, SLAC) at Stanford University, has received a National Institutes of Health (NIH) grant to develop and integrate high-throughput robotic technologies to greatly accelerate the discovery of the 3-D structure of proteins, providing key insights about their biological function.
Ad Emmen [News on Advanced IT] [Calendar] [Analysis] [IT in Medicine] |
 |