fb-pixelComputers may predict drugs’ side effects - The Boston Globe Skip to main content

Computers may predict drugs’ side effects

Cambridge lab aims to detect potential problems before tests on humans begin

Scientists at a biomedical research laboratory in Cambridge have developed a computer model to test drugs for adverse side effects before they would be given to people in clinical trials. If proven successful, such an approach could be a significant step toward solving a major problem facing the pharmaceutical industry.

Adverse drug reactions are a costly liability, causing about 20 percent of drugs in clinical development to fail, according to the Tufts Center for the Study of Drug Development. The ability to detect them early in the development process could spare patients harmful reactions and save the time and expense of launching clinical trials on drugs that carry problematic side effects.


Already, companies test drug candidates in animals and in the lab to minimize the possibility that drugs with toxic effects are tested in people. But those offer only a limited way to mimic how a drug will act in humans, so crafting tests that could flag molecules likely to have side effects has become been a priority.

“I think this is an extremely powerful tool,” said Kenneth Kaitin, director of the Tufts Center for the Study of Drug Development, who was not involved in the study. “One of the major challenges for the industry is that compounds enter clinical testing without a very good understanding of how the drugs work and what the side effects are going to be, so you’re essentially entering blind.”

The team of scientists, from the Novartis Institutes for BioMedical Research in Cambridge and the University of California San Francisco, focused on drugs’ “off-target” effects. Drugs are usually designed to do a particular job in the body, but they also interact with a person’s biology in complicated and sometimes unpredictable ways. For example, the weight-loss drug fen-phen was withdrawn from the market because of patients’ deaths. Scientists later discovered the biological reason: In the body, the drug also activates a separate biological pathway, leading to heart disease.


In the study, published Sunday in the journal Nature, the researchers started by creating a database of 73 proteins that were associated with adverse side effects. They used a computer program that could rapidly analyze drug molecules to predict whether a library of 656 approved drugs were likely to interact with any of the proteins linked to side effects.

They made more than 1,000 predictions and verified them by doing extensive searches of published materials and conducting new tests in the laboratory. About half of the predictions were correct, and 151 of the interactions were previously unknown.

For example, they found that a prostate cancer drug had an unexpected off-target effect. It acted in a similar way to a class of painkillers that includes aspirin.

Those drugs, called nonsteroidal anti-inflammatory drugs, can cause problems with the gastrointestinal system, which helped explain a previously mysterious side effect of the prostate drug: upper abdominal pain.

The technique is already being used at Novartis.

“The beauty of computer programs [is] they are very inexpensive,” said Eugen Lounkine, a computational biologist at Novartis who led the work. “They don’t need the actual physical compounds, so they can assess thousands of those virtual compounds, and that allows us to very, very early on prioritize what compounds we want to look at in further experiments.”

Kaitin said the company’s publication of the results marks an interesting shift in the pharmaceutical industry, which has been more protective of its tools and approaches.


“Safety of patients is of global interest and a huge ethical issue,” said Laszlo Urban, global head of preclinical safety profiling at Novartis. “It gives the scope of what can be done, by any company, to make a safer and faster way to the clinic.”

Carolyn Y. Johnson can be reached at cjohnson@globe.com. Follow her on Twitter @carolynyjohnson.