Advanced Architectures for Mass Spectra Deconvolution, 10-R8012

Printer Friendly Version

Principal Investigators
Matt S. Argabright
Keith S. Pickens
Jonathan A. Bohmann
Will L. Arensman
Pamela B. White

Inclusive Dates:  01/05/09 – 07/14/09

Background - The objective of this effort was to apply advanced multi-core computing architectures to mass spectrometry (MS) spectra deconvolution. MS spectra deconvolution and compound identification is a computation-intensive problem with applications in food safety, scientific research and national security.

Approach - The problem was divided into two segments. The first segment addressed algorithmic risk, and the second segment addressed platform risk. The algorithmic development was carried out on a commercial four-core processor. This provided a rich and full-featured development environment that maximized the efficiency of the parallel algorithm development and coding. The second segment of the program focused on a highly parallel platform with hundreds of processing elements. Because the algorithm was stable and well understood, the port progressed rapidly; therefore, the primary focus was on the issues unique to the platform.

Accomplishments - During the first program phase, two code bases were successfully implemented with the anticipated enhanced performance from the parallel architectures. One code was an SwRI algorithm for deconvolution. The second was a publicly available code base for mass spectra analysis. In the second phase of the project, the port of these algorithms to highly parallel architectures was undertaken. This phase was successful in defining the limitation of the platform for these types of problems. While a hypothesized 10x speedup was not attained using General Purpose computing on Graphics Processing Units (GPGPU) technology, a 4x and linear speedup was attained through the use of Open Multi-processing (OpenMP). These results provide valuable skills and insight for guiding future parallelization efforts.

2009 Program Home