Merchants of False Hope

Francis Collins and J. Craig Venter are the scientists whose names will always be associated with mapping of the human genome — after racing to complete the job, the two announced in 2000 that it was done, three years early.

So it’s impressive that this year they both announced another shared discovery. The two scientists separately sent samples of DNA to companies that sell personalized genetic reports — the firms that offer “your risk analyzed for 118 diseases and traits,” as one of them, 23andme, puts it.

If personalized genomics really could reveal “your genetic predisposition for important health conditions,” as Navigenics’ website says, then the different companies’ reports should match: When 23andme says Francis Collins’ risk of getting adult-onset diabetes is high, then a Navigenics analysis should tell him the same thing.

But the companies (those two in particular, and others) did not agree with each other. In a speech last June, Collins said (apparently without naming the firms or the specific illness involved) that his tests came back sounding like a genetic version of Goldilocks and the Three Bears: One firm said he was at “high risk” for a disease, another said he was at “low risk,” and the third told him his odds were average.

Then, in the October 7 edition of Nature, Venter and his colleagues Pauline C. Ng, Sarah S. Murray and Samuel Levy reported the same experience. They sent five individuals’ DNA samples to Navigenics and 23andme for analysis and looked at predictions of risk for seven important diseases. Fewer than half the predictions agreed.

Collins and Venter’s statements are a wake-up call for a new industry, which has rightly been greeted with skepticism from many scientists and regulators. Among several important suggestions, for instance, Venter’s Nature paper calls on firms to be more transparent about how they calculate risk, and to standardize the genetic markers they choose to use in their analysis.

For the moment, personalized genetic tests are more akin to pseudo-science than real science. That’s not because the testers can’t agree (disagreement is a normal part of any scientific enterprise). It’s because of the reason they disagree: Much remains unknown about (a) how genes work in general, and (b) how multiple genes are involved in the onset of a particular disease and (c) how genetic factors interact with a person’s environment and behavior and (d) whether people who know their risk profile will in fact change their behavior (that’s important because the risk of diseases like heart ailments, diabetes and stroke depends in part on how people act).

With all four issues still open for broad discussion, each genetic testing firm makes its own judgment calls about what the data mean. As Venter and his co-authors argue, there’s nothing wrong with that, as long as the companies come clean about it, and show their work, as we used to say in math class.

If they do, personalized genetic testing could become a teaching model for explaining how scientists actually estimate risk. They don’t do it by claiming sure and certain knowledge of the future. They do it by using current knowledge (while admitting that the state of that knowledge is changing) and applying current theories (knowing those too will change) to arrive at a guess, and they tell you how probable they believe their guess to be, and why. The result isn’t “Science says eat more bran!” It’s more like “eating bran is associated by this particular method with longer life in these people.”

That would not be as easy to market, but it’d be much better for society. After all, to address any global problem, from climate change to food security to nuclear proliferation, we need to understand how to estimate risk. It’s a societal problem, then, when corporations sell the science of risk in a way that encourages people to misunderstand it.