Using clouds for data - intensive computing in proteomics

Abstract Type: Position Proteomics is an integral part of systems biology research and one that has become largely a dataintensive field. Due to advances in proteomics technologies, mass spectrometry data archives have been on the rise, as are sequence data generated from next-generation sequencing technologies. However, data mining and applying new methods to existing data sets is currently limited by the need to move and manage large sets of data. For example, mining a proteomics archive of MS/MS spectra for post-translational modifications is currently done on only small subsets of data, not the entire archive which contains over a billion MS spectra occupying over 150 TB as of September 2009. The problem is compounded when trying to co-analyze proteomics archives and nucleotide data from a broad spectrum of sequenced organisms – a capability needed when dealing with environmental samples. Application of the methods is limited by the manual labor required to gather all the data on a single file system. Yet integrated analysis of data from different sources has become far more important for furthering discovery. Cloud computing is expected to be an ideal computing model that can address these limitations and support the massive scale integration and analysis of proteomics and genomics data in a way that is transparent to the application scientist. In this presentation we will identify the different ways & associated challenges that will be faced while developing cloud computing frameworks for proteomics applications. For ease of exposition, we will organize the presentation from the perspective of two types of core compute operations that are prevalent in proteomics & metaproteomics analysis: i) Database search – the primary example of this application class is peptide identification from MS/MS spectral data; and ii) Large-scale graph analysis – the primary example of this class is protein family characterization which typically involves computational of large-scale all-against-all sequencesequence, profile-profile and sequence-profile comparisons. Both these applications are central to the structural and functional characterization of proteins represented from single organism to more complex microbial communities. These applications harbor a high potential to benefit from cloud computing because of large sizes of data and a broad user-base. Yet, their portability into the paradigm throws several challenges, both algorithmic and systemic. Historically, the codebase for these applications have been serial and there are very few implementations that support even traditional parallelization (i.e., on distributed and shared memory machines). Therefore, algorithmic innovations are required to map the underlying problem space to a Map-Reduce model, which is becoming the de facto standard for cloud computing. Furthermore, the evolution and continued upgradation of cloud computing technologies and infrastructures, at both the architectural and system software levels, is poised to impede transition and wider adoption. To jumpstart discussion, we will present ideas and preliminary findings of our on-going research in the area. The overarching goal is to identify the merits and current limitations that exist along the path toward implementing the grand vision of building cloud computing infrastructures for proteomics research and applying them toward transformative advancement of scientific discovery.