Â鶹´«Ã½

IT IS a disarmingly simple idea: to find out if a drug might treat a disease it wasn’t intended for, check out whether it has an opposite effect on gene activity to the illness itself. How do you find such drugs? By mining large public biological datasets.

For more than a decade, so-called DNA chips have routinely measured the activity of thousands of genes at a time, and researchers have deposited the results online into the , after their papers were published.

, a bioinformatician at Stanford University in California and colleagues reasoned that it should be possible to find new drug uses by combining data from GEO with information gleaned from another database – the . In this database, biologists at the Broad Institute in Cambridge, Massachusetts, have documented how patterns of gene activity in human cells change when they are exposed to a range of drugs.

Butte’s team mashed up the two datasets according to a simple hypothesis: drugs that have an opposite effect on gene activity to a particular disease could be good candidates for treating the condition. So the researchers devised algorithms to look for drugs that ramp up the activity of genes that are unusually quiet in tissues affected by a particular disease, and suppress those that are hyperactive in that disease.

Butte admits that colleagues doubted the GEO data would be good enough to provide valuable insights. “When people see something that is free and on the internet, they think it must have no value,” he says.

“When people see something is free and on the internet, they think it must have no value”

But Butte’s team proved the sceptics wrong by taking two of the strongest leads and showing in animal experiments that the drugs could treat the conditions with which they were paired. In one case, the epilepsy drug helped rats with (Science Translational Medicine, ); in the second, , used to treat stomach ulcers and acid reflux, reduced tumour growth in mice implanted with human lung cancer cells(Science Translational Medicine, ).

“This shows that simple, elegant ideas really can come through,” says , also at Stanford, who has used data mining to find combinations of drugs with dangerous side effects (Â鶹´«Ã½, 4 June, p 16). “Researchers will be saying to themselves, ‘Why didn’t I think of that?'”

One snag is that patents on the two drugs have expired, so firms won’t have the financial incentive to run clinical trials to find out if the new uses are viable. But the same approach could also highlight multiple uses for drugs still in development.

“This is a technique that’s very promising,” says , director of computational biology at GlaxoSmithKline in King of Prussia, Pennsylvania.