Parallelization of the spectral deconvolution stage of the proteomic discovery process

Mass spectrometry is one of the primary analytical tools for the quantification and identification of proteins in human body. Scientists analyze the data generated from mass spectrometry instruments to detect protein patterns. The operations performed on mass spectrometric output to detect such patterns include spectrum visualization, deconvolution, alignment, normalization, statistical significance tests, and pattern recognition. Bindley Bioscience center in Purdue University's Discovery Park hosts a Omics Discovery Pipeline (ODP). The ODP is available as a web-based analysis platform that performs the aforementioned operations.

The time taken to generate output in the spectrum deconvolution stage is a major concern. Much of this concern results from the fact that this stage has to deal with a sizable number of large input files that are present in potentially many different input formats. Due to slow operation speed of this stage, overall performance of the process for protein quantification and identification is hampered. Our research work concentrates on discovering and implementing ways to increase the performance of the spectral deconvolution stage of the process for protein quantification used in the ODP by leveraging multi-core and multi-node architectures. By leveraging OpenMP and MPI, we were able to obtain an order of magnitude speedup of the spectrum deconvolution stage of the ODP.