News Archive

New Computational Technique Helps Predict Drug Side Effects

Published 12/19/2007

Early identification of adverse effects of drugs before they are tested in humans is crucial in developing new therapeutics, as unexpected effects account for a third of all drug failures during the development process. Researchers at UC San Diego (UCSD) and the San Diego Supercomputer Center (SDSC) at UCSD have developed a novel technique using computational modeling to identify potential side effects of pharmaceuticals, and have used the technique to study a class of drugs that includes Tamoxifen, the most prescribed drug in the treatment of breast cancer. Their study is currently available on line at PLoS Computational Biology.

Conventional test methods screen compounds in animal studies in advance of human trials in the hope of identifying the side effects of promising therapeutics. The UCSD and SDSC team - led by Philip Bourne, Ph.D., professor of pharmacology at UCSD's Skaggs School of Pharmacy and Pharmaceutical Sciences and SDSC Structural Bioinformatics Lead, and SDSC researcher Lei Xie, Ph.D. - instead uses the power of computational modeling to screen specific drug molecules using a worldwide repository, the Protein Data Bank (PDB), which contains tens of thousands of three-dimensional protein structures.

Drug molecules are designed to bind to targeted proteins in order to achieve a specific therapeutic affect, but if the small drug molecule that functions as a "key" attaches to a different off-target protein that has a similar binding site, or "lock," then side effects can result.

To identify which proteins might turn out to be unintended targets, the researchers take a single drug molecule and search for how it might bind to as many of the proteins encoded by the human proteome as possible. To illustrate the novel approach, in this published case study they looked at Select Estrogen Receptor Modulators (SERMs), a class of drug that includes Tamoxifen.

"The computer procedure we developed starts with an existing three-dimensional model of a pharmaceutical, showing the structure of a drug molecule bound to its target protein; in this case, the SERM bound to the estrogen receptor," said Bourne, co-director of the Research Collaboratory for Structural Bioinformatics (RCSB) PDB, which is housed at SDSC. The scientists then use computer analysis to search for other binding sites that match that drug binding site - like looking for other locks that can be opened by the same key.

In this study the researchers screened the SERM drug binding site against some 800 human drug-related structures in the PDB. In the process they used the GEMSTONE (Grid Enabled Molecular Science Through Online Networked Environments) software developed in a collaboration including SDSC.

In their results, the team found a previously unidentified protein target for SERMs. The identification of this secondary binding site explains known adverse effects, and opens the door to modifying the drug in a way that maintains binding to the intended target, but reduces binding to the second, unwanted site.

"If a drug has adverse side effects, it is likely that drug is also binding to an unintended, secondary molecule; in other words, the key that allows it to attach to its target fits more than one lock," said Bourne. He explained that using this computational technique to find another "lock" could result in one of three things: the new lock might show no effect; the lock could explain an adverse side effect of the drug; or the research could potentially discover a new therapeutic effect for an existing drug - drug repositioning.

The UCSD researchers are continuing their studies, which Bourne says can be applied to any drug on the market for which a structure of the drug bound to the receptor exists in the PDB. Bourne emphasized that results from this approach still need to be tested experimentally.

In the future, SDSC's Xie explained, it will be useful to develop a service that makes this screening capability available through a Web interface where researchers can submit potential drug molecules and quickly identify possible interactions. To return results in a reasonable time will require the power of SDSC supercomputing resources to be able to screen submitted molecules against approximately 20,000 non-redundant structures in the PDB.

Jian Wang of UCSD's Bioinformatics Program also contributed to the PLoS study. The work was supported in part by the National Institutes of Health.

Related Links

The Bourne Laboratory
RCSB Protein Data Bank (PDB)
San Diego Supercomputer Center (SDSC)
PloS Computational Biology