Predicting Protein Complexes

A combination of genomics data and molecular dynamics modeling is sufficient to predict protein complex structure

The zone where two proteins interact presents a possible target for drug design. But identifying possible drugs requires a detailed understanding of the interface between the proteins. Computer simulation provides a useful tool for gaining such an understanding. But simulating protein complexes can be challenging, especially when the interactions are fleeting—such as when signaling molecules attach and detach in a flicker. Now, a new method can efficiently predict the structures of transient protein complexes from a combination of genomic and structural data.


“This is an entire approach to protein-complex structures based on several different computational methods,” says Hendrik Szurmant, PhD, coauthor of the paper and an assistant professor of molecular and experimental medicine at the Scripps Research Institute. The work was published in Proceedings of the National Academy of Science in December 2009.

This complex of membrane-bound sensor histidine kinase (TM0853, in blue) and its response regulator (TM0468, in red) shows the contact residues identified from genomic data in orange and blue. The catalytic residues that exchange a phosphoryl group are shown in yellow. Courtesy of Alexander Schug and Hendrik Szurmant.

To determine the structures of proteins in complexes, researchers have used both homology modeling and purely physics-based molecular dynamics simulations. But both approaches have proven less successful than hybrid approaches. Szurmant and his colleagues developed their new hybrid approach using a two-component signaling system in bacteria as a test case. The system consists of a membrane-bound kinase that passes a phosphoryl group on to its response regulator within the cell. The team analyzed databases of genomic sequences for almost 9000 examples of the two co-evolving proteins looking for co-varying mutations. Their aim: to identify likely points of contact between the two players, under the theory that one protein’s contact residue tends to match a mutation in its partner. Then they used those points of contact to combine the two proteins in a molecular dynamics simulation.


“Our method brings the two proteins close together in a computationally very inexpensive way, then as a very last step the structure is refined in a molecular force field,” says Szurmant. The combination of approaches minimizes computation time, he says, compared to methods that rely more heavily on molecular dynamics.


When the researchers tested their methods on a complex whose structure had already been determined, the prediction was in excellent agreement with the known structure. They also tackled a then-unresolved complex from Thermotoga maritima, TM0853/TM0468. An x-ray diffraction structure of that complex has since been published, confirming many aspects of the prediction.


This technique could be used for other types of systems, says Szurmant, so long as enough sequence information is available for the genomic step to pick out statistically significant variations. “The approach relies on variability, so if the system is very conserved, one would need a lot more sequence,” he says. The team’s next step is to apply the method to other bacterial systems, and eventually to develop an online tool to make the approach available to other researchers.


This work shows that the combination of genomics data and molecular dynamics modeling seems to be sufficient to predict protein complex structures, says Angel Garcia, PhD, professor of physics at Rensselaer Polytechnic Institute. Garcia points out that the accuracy of the method is particularly impressive. He adds, “I think almost anyone that is working on a given complex is going to try this for their own pet system.”

Post new comment

The content of this field is kept private and will not be shown publicly.
This question is for testing whether you are a human visitor and to prevent automated spam submissions.
Enter the characters shown in the image.