by Emily Carlson
November 29, 2004
Everyone enjoys a little friendly competition, including protein structure prediction researchers. In fact, more than 200 groups of these experts worldwide recently created some 15,000 different computer models of protein structures to see who could make the best predictions.
Successful modelers will present and discuss their work in Gaeta, Italy, from December 4-8 at the Critical Assessment of Techniques for Protein Structure Prediction meeting (CASP), sponsored in part by the National Institute of General Medical Sciences (NIGMS), a component of the National Institutes of Health (NIH). The meeting, along with a community-wide experiment that precedes it, takes place every other year.
“The CASP experiment and meeting provide a very good mechanism for scientists to come together and critically assess the methods they use to model the structures of proteins,” said Jerry Li, Ph.D., an NIGMS program director in protein research and bioinformatics who also heads the institute’s involvement in CASP.
Instead of conducting wet lab experiments to determine the three-dimensional structures of proteins, molecules made of amino acids, protein structure prediction researchers use simulations. The models are based on either similar proteins whose shapes have been determined experimentally or on the most likely arrangement of atoms.
Currently, it normally costs over $200,000 and takes months to years to solve a single protein structure experimentally, said Li. He added that taking advantage of ever-improving computer power and better modeling methods could drop the cost to less than $100 and the time to less than an hour.
Toward this end, CASP participants submitted their prediction models for selected proteins, along with abstracts describing the methods they used to arrive at those models, from last May to August. The modelers did not know the structure of these proteins. Only the independent reviewers, who judged the accuracy of the predictions, were privy to the actual protein structures that had just been determined experimentally but not yet published. In November, the reviewers invited the groups with the most accurate models to present their work at the CASP meeting.
At the 2002 meeting, the best predictions were off by several angstroms—just a few ten-billionths of a meter, said Li, adding that although this number is small, it isn’t small enough. “If we want to use protein structure prediction for drug discovery, we’ll need even higher accuracy,” he said. “We want to push the model accuracy to within 1 angstrom.”
This is where the CASP meetings fit in. As Li explained, no critical assessment of the methods and technology used to predict protein structures existed before the first CASP meeting in 1994. Since then, the group has gathered to learn from each other, work together, discuss progress in problem areas, and set future goals. As a result of the meeting, modelers often form new academic and industry collaborations, and successful methods are recognized and adopted by other researchers.
These advances in protein structure prediction could lead to a better understanding of protein functions and rational drug design, ultimately making everyone a winner of the CASP experiments.
The meeting is co-sponsored by NIH’s National Library of Medicine, the Istituto Pasteur-Fondazione Cenci Bolognetti, the European Molecular Biology Organization, and the BioSapiens Network of Excellence.
|
