Feature - Protein structure: taking it to the bank

At left, the predicted (gold) and experimental (blue) structure of 30S ribosomal protein S27A from Thermoplasma acidophilum. At right, the same for protein G. Xu′s team participated in the 2008 Critical Assessment of Structure Prediction (CASP) competition, sponsored by NIH, BioSapiens Network and the European Molecular Biology Organization, in which participants are given 120 proteins with unknown structures to predict. RAPTOR ranked among the top 5 of 85 teams.  Thermoplasma acidophilum, represented above, was a test protein used in CASP. Image courtesy of Jinbo Xu.

Properly functioning proteins are essential for our bodies. A protein’s structure, the folded form its amino acid string assumes, determines its function.  Scientists know the sequence and structure of about 50,000 proteins—out of millions. They keep this valuable information in a “bank”—the NSF- and NIH-funded Protein Data Bank.  To predict the structure of a newly identified protein, scientists can compare it to a similar banked protein. 

This works quite well, but what if there is no similar protein in the bank?  Then it’s back to first principles: creating predictions from scratch using the physical principles that describe the interactive forces between atoms.  Experimental methods, such as x-ray crystallography and nuclear magnetic resonance are time-consuming and costly, and cannot be used on all proteins, so scientists have turned to computational predictions.

RAPTOR is short for Rapid Protein Threading Predictor Image courtesy of bioinformaticssolutions.com.

Researchers at the Toyota Technological Institute at Chicago, an affiliation of the University of Chicago, are using the Open Science Grid to quickly and accurately simulate protein structures, and thereby to determine their functions.

An average protein contains about 250 amino acids, each of which has at least 10 atoms. To develop a prediction, scientists must determine the position of the atoms in the system. Each atom has many possible positions. Scientists must search all possible conformations to find the most stable one—a computationally demanding process, says Toyota Technological Institute researcher Jinbo Xu.

RAPTOR on OSG: quick and accurate

Xu and his team use RAPTOR, a molecular modeling software package developed by Xu, to run thousands of “small protein” (under 100 amino acids) simulations on OSG. RAPTOR, available to any researcher, samples all possible configurations against their mathematically estimated probability of being stable. They evaluate the results to find the most stable, hence the most likely to be the “true” structure.  Xu’s team assures RAPTOR’s accuracy by predicting experimentally determined protein structures and comparing the predictions with the known structures.

The simulations are independent of each other, so OSG’s distributed framework is ideal for running them quickly and simultaneously.

“OSG has helped us shorten our folding simulation experiments from months to days and is now an essential computing platform for our research,” Xu says.

—Amelia Williamson, for iSGTW