March 8, 2017
The basketball court isn’t the only place that Purdue has been winning lately. Using the Carter and Rice community cluster supercomputers operated by ITaP, Professor Daisuke Kihara and his team have proven themselves among the best in the world at modeling protein structure and interaction.
Kihara, a professor of biological sciences and computer science, and his team – graduate student Lenna Peterson, visiting professor Genki Terashi, and postdoctoral researcher Woong-Hee Shin – achieved top scores in two recent contests: CASP (Critical Assessment of protein Structure Prediction), in which participants predict the three-dimensional structure of a protein based on its amino acid sequence, and CAPRI (Critical Assessment of PRediction of Interactions), in which they predict how two or more proteins will dock together to form protein complexes.
At the same time Kihara and other bioinformatics researchers were trying to model the proteins computationally, structural biologists were using techniques such as X-ray crystallography and nuclear magnetic resonance spectroscopy to determine the actual shape of the protein targets, and teams were scored based on how well their predictions matched the experimental results.
Kihara’s team was one of a handful of top finishers chosen to present at a joint meeting to assess the results after the competition. They were among the top groups in the model refinement category in CASP. They were also the top group in the CAPRI category known as “scoring,” in which participants must choose the best out of about a thousand models of protein complexes provided by the organizers. “Computational models always give you lots of results,” explains Kihara. “The difficult part is how to score them and pick the right one.”
Protein structure prediction generally relies on two methods. In template-based modeling, models are created based on similarity to known structures stored in the Protein Data Bank, which contains over 100,000 solved structures. If there’s not enough similarity to a known structure, researchers have to use the “de novo method” and build the models from scratch. Because the de novo method is so difficult, most researchers, including Kihara, use a hybrid of the two methods, taking parts of the structure from similar sequences in the database and figuring out how to assemble them.
Kihara used many core hours on the Carter and Rice clusters to run molecular dynamics simulations that compute the force between every pair of atoms and the movement of atoms and molecules. By simulating how the protein moves under different conditions, researchers can test and refine their structure models.
Access to Purdue’s Community Cluster Program supercomputers and their processing power proved critical to success because of the number of simulations that need to be run for each protein target. “We run many different simulations with different parameters to see how they converge. If all the simulations point to the same structure, then that’s a very good indication that the structure is correct,” says Kihara.
It’s not just one protein target that the researchers had to model. Over the course of the several month-long competition, new protein targets were regularly released and had to be solved within weeks, meaning that Kihara’s success was the result of a tremendous human and computational effort.
To learn more about Purdue’s Community Cluster Program, contact Preston Smith, ITaP’s director of research services and support, email@example.com or 49-49729.