Datawars: Grid Computing Democratizes Proteomics

Erica P. Johnson The young field of proteomics has quickly risen to match physics and meteorology in its huge appetite for computational capacity. According to Sylvie Langevin, development group manager of Montreal-based proteomics firm Caprion Pharmaceuticals, the computer processing and data handling requirements of proteomics exceed those of genomics by a factor approaching one million. "In genomics we have files of several kilobytes in size, but in proteomics we're closer to 1 [gigabyte]," Langevin says. Fortunately proteomics problems, such as three-dimensional visualization of protein structures and searching for matching sequences, can readily be broken up into small components that can be processed separately on different computers and then reassembled at the end. "In the trade we call them 'embarrassingly parallel,'" says Brian Carter, head of life science solutions for IBM in the United Kingdom. GRID LINES This makes proteomics ripe for grid computing, the fast-emerging computational model for sharing resources such as servers and databases within and between enterprises. A data grid--whether internal or external--provides a single point of access to multiple data sources. Computational grids break down a single problem, such as a sequence search, so that it can be processed in parallel even by as many as several thousand computers, in extreme cases. Using grids to access public databases will likely be particularly appealing to proteomics firms that now experience difficulties identifying relevant data within the large web of external sources. This leads to another key point concerning human resources, which is that effective exploitation of grid computing requires a synthesis between information technology (IT) and proteomics skills. Assembling such a team will be beyond the budget of many smaller firms, which may therefore need to rely on service providers (such as IBM) that already have the requisite skills. Caprion Pharmaceuticals is a typical case, dedicating four large servers, each with 16 to 24 gigabytes of main memory, to proteomics problems, and reserving a fifth server for development of new software. In common with other proteomics companies, though, Caprion is precluded from using an external grid by intellectual property protection constraints. According to IBM's Carter, companies are deterred not just by the fear of valuable data being stolen, but also by the fact that legal intellectual property entitlements are lost if relevant data are transmitted on public networks. In April, IBM plans to announce a new service, initially available just in the United Kingdom, aimed at proteomics and biotech firms, which will provide access to secure computing power. The company plans to expand globally as well. PUBLIC SOURCES Data grids can assist in collaborative projects, by making communal data available to all participants while protecting intellectual property rights and confidential information. This is relevant for the London-based drug discovery company Inpharmatica, which is in a partnership with the leading European biotechnology company, Swiss-based Serono, to identify novel therapeutic proteins. Inpharmatica already uses a computational grid internally for predicting the structure of proteins based on the known underlying genomic sequences, and the company will want to exploit this in the joint project. Although data and computing grids have obvious potential, there are alternative ways of solving proteomics IT problems. "Most of our common informatics challenges involve building cooperative information tracking systems, integrating heterogeneous data, and developing algorithms that address bottlenecks better," says Thodoros Topaloglou, vice president of informatics at MDS Proteomics,1 one of the pioneering proteomics companies. Apart from security and control of access to both computers and data, the main prob0lems with grid computing are: allocation and management of resources so that computation can take place within a predictable time frame; and ensuring that data is in the right place at the right time. In an external grid comprising servers that potentially could be used by anyone in the global research community, it is impossible to predict the exact duration of a particular proteomics process unless there is clear information about availability of resources at a given time. "A big part of the grid standards drive is focused around the directory definition of resources," notes Bob Rosenthal, a consultant on IBM's grid computing team. These open-grid standards are being defined by the Global Grid Forum, and will gradually take over from proprietary protocols, which predominate at present. The standards also will embrace security, in particular, for regulating the use of and access to specific servers and resources, and for defining priority users. The future of grid computing now seems assured in the case of proteomics, for it is the most affordable way of satisfying the field's hunger for processing power and data storage capacity. Philip Hunter (phunter@compuserve.com) is a freelance writer in London. 1. A.J.S. Rayl, "On the fast track in functional proteomics," The Scientist, 15[7]:10, April 2, 2001.

Datawars: Grid Computing Democratizes Proteomics

Interested in reading more?

Become a Member of