Monday, January 6, 2014

The problems with genomic prediction. Need it be stated again?

We often write about the challenges to prediction of phenotypes, especially disease, from DNA data.  If the trait is clearly due to variation at a single gene, then prediction can be useful.  If most instances are due to variation at a single gene, but not always the same gene, DNA data can be useful as well.  Of course if the variation is in regulatory or other DNA, rather than protein coding parts (genes proper), unless the whole genome is sequenced (not just the 'exomes'), things may be much more unclear.

Most common disease or behavioral traits aren't like that, however.  As we have piles of data to show, traits like cancer, stroke, dementia, and heart disease are typically due to very many but individually minor, contributing parts of the genome, plus, oh yes!, maybe some Environmental factors that, if we're geneticists, we don't really want to have to think much about)  But the big E is the key to trouble, in ways reflected by two stories in yesterday's NY Times Sunday Review section.

The cancer example
One story suggests what is presented as a surprise, that cancer (despite the 40-year 'war' against it by NIH) is rising in lifetime risk, not falling, or at least is going to affect more rather than fewer people in the future.  But, despite the fact that the 'war' on cancer was a bad way to waste a lot of money (good proposals would have been funded without the attraction of a poorly disciplined pot of gold), even if cancer treatments are better (sometimes true, at least), more victims will fall to cancer.  This seems counter-intuitive, so why?

The answer is something well-known to those who know, but poorly explained to the public.   We've written about it before here on MT.  It's called 'competing causes'. As long as we all have to die of something, and the older we are the more damage our various cells sustain, if you remove one cause, those who are spared that then stay alive to be vulnerable to the remaining causes.  Heart diseases can strike relatively early, but cancer, which in many if not most instances is due to mutational and other cellular damage, is a risk that keeps on growing as cells continue to be exposed to mutagens and the like.

This is no new discovery or surprise and the public and all geneticists should by now be well aware of it.  Also, earlier treatment that is developed for other diseases will have similar negative consequences (though of course to those who were saved from earlier death).  The overall health care cost burden is likely the result of research that is successful against earlier-age killers.

Other diseases are part of the Environment of cells unaffected by those diseases. Spared of those, you are exposed to continued mutagens throughout your life.  Yet these changes cannot be predicted, and as a result, genotype-based risk estimates (that aspect of 'personalized genomic medicine') are a fantasy: they may or may not be right and are of unknown--and unknowable--accuracy.

Socioeconomic advances
The other Times article yesterday that caught our attention was about the increased risk of various diseases to those who work their way out of the lower socioeconomic strata in our society.  As they rise in education, wealth and so on, and perhaps because of the stress required to dig oneself out of poverty, people suffer risks of various diseases that are higher than they would have experienced or of their peers in their newly achieved SES group.

Again this is the mixed blessing of changing lifestyles.  It is the again-unpredictable effect of Environmental change.  It's unpredictable even if we knew the specifics of a given sample of cases, because they are always retrospectively analyzed, after things have already happened to that sample, but we can't know how things will change in regard to SES-specific risks or in SES-mobility patterns in the future, the future of those who want to know their risks.  You cannot tell these things from the person's genotype, of course, and social science is even farther behind genomics when it comes to having a crystal ball for the future.  Indeed, all the Big Data resources in the world will have this problem, and to that extent will be a huge waste of public health funding.

These are not new major discoveries of new principles.  They just happened to be in the paper and to reflect the issues.  They just sound the warning bells that should be being heard, and heeded, in the labs and offices of NIH and medical schools, and the companies who have been giving what is essentially misleading (and inconsistent) DNA-based health advice, the whole community making rather careless promises of genome or other 'omic' based miracles.  These are facts of life, so to speak, that in themselves should be the subject of investigation. But that's a real, legitimate challenge, unlike the weak challenge of proposing to collect more and more data without being obligated to deliver on promises made to justify that or to address the real problems in an effective way.

16 comments:

Jim Wood said...

I wondered if you'd seen the cancer op-ed piece, Ken, which "announced" things you've been saying for years. Nothing surprising in that one. The other about health risks among achievers born into poverty was more surprising -- at least until I thought about it, after which it made perfect sense. To quote pertinent platitudes: "There is no free lunch." Or, as the immortal (i.e. dead) Tennessee Ernie Ford once sang regarding competing causes, "If the right one don't getcha, then the left one will." There seems to be a strong human propensity, perhaps ancient, perhaps modern, to want to "cure" aging and death, despite the absence of one jot of evidence that either is possible. To quote another sage (Woody Allen), "I don't want to become immortal through my art; I want to become immortal by not dying." Perhaps the only modern spin on this desire is that we believe salvation will come in the form of science and medicine instead of divine intervention. Both are matters of pure faith.

Ken Weiss said...

Yes, and the point is that one can't predict what that future not-free lunch will be. So why do people put any faith in predictions-from-birth as in DNA sequence?

One an argue for some such predictions (known high-penetrance variants), but perhaps what is needed is a new approach that works with age-specific hazard functions, and a more sophisticated concept of 'risk'.....not to mention having adequate understanding of genomic causation, and a public health policy that accepts that nature of aging and mortality.

Or, for those seeking immortality, have them drop their money in the basket in church rather than in 23andMe's basket.

Manoj Samanta said...

"the 'war' on cancer was a bad way to waste a lot of money (good proposals would have been funded without the attraction of a poorly disciplined pot of gold) "

I call it modern day version of medieval alchemy, where science was driven by how to turn mercury into pot of gold.

Jim Wood said...

Some demographers used to model age-specific mortality risk with hazards functions that included gamma or lognormal distributions for "unobserved heterogeneity" to adjust for all the stuff we don't know about, including *perhaps* genetic variation. Granted, this approach didn't tell us anything about the causes of the heterogeneity, but at least it had the virtue of intellectual honesty. Now most of the same demographers are looking for single-gene variants that account for large fractions of the risk. A major step backwards, in my opinion.

Ken Weiss said...

I think that 'war' was based on the success of the Manhattan project (if an A-bomb can properly be called a 'success'). But even given that, 'cancer' was never a focused problem with an appropriate underlying theory.

Again, good cancer proposals would have been funded, and a lot of junk ones not, had the funds just been in the usual RO1 kitty. But this further set the precedent for Big Data proposals and projects as they are now so cutely termed.

There are several such endless projects already draining funds from the NIH pot, and new ones being proposed all the time (brain map; various genomic data bases; long-term epidemiological studies that shall be nameless....)

Ken Weiss said...

It's worse than you say, even. First, these 'demographers' are hungry to apply genetics to all sorts of behavior--dipping into the eugenics pot again. And second, even doing as you say will only address exposures retrospectively and tell us about 'risk' in the past, that's already been realized. Nobody wants to face up to the clear fact that environmental factors, broadly meant, cannot be predicted with accuracy (as at least the second, SES-based Times story implicitly showed).

Holly Dunsworth said...

I will not even try to match your poetry, Jim and Ken, but I want to add that maybe our obsession with determinism isn't as much about avoiding death but about gaining the license to be naughty (or at least behave less than healthy) before we get to the inevitable end.

Ken Weiss said...

I want to fund research
To find the perfect shots
That keep me on my perch
So I can drink _my_ shots!

Holly Dunsworth said...

:) I meant, to identify if folks who may want to drink to they die are at higher risk of things so that they reconsider their behavior. Otherwise, for everyone else without those hypothetical genetically determined risks: party time!

Ken Weiss said...

Right, but what people want is vaccines or pills to cover everything so they can smoke, eat, drink, and be merry. The don't want to know that they are genetically risk free, or at least, they just want reassurance they can get a pill for whatever they want to do.

So, one might say, the way 'the pill' changed sexual behavior, everyone wants 'the pills' that let you smoke and drink away!

Holly Dunsworth said...

I'm really out of it, a real alien, if that's indeed what people want. I never heard of such a hypothetical! I just thought that in terms of genetic determinism of disease risk, people were looking to those folks, for example, who smoke til they're 100 and hoping that if only they could somehow now that they have the same (whatever that is) biology, then they wouldn't have to quit smoking either, etc etc...

Holly Dunsworth said...

*know* (I'm not on the ball today)

Ken Weiss said...

It's just that when I've given general audience talks, the question afterwards is in terms of when will we have a pill that will allow them to smoke. And if you turn out to be, say, hypersensitive to alcohol, tobacco, marijuana, or sex, then will you quit those things? I wonder how many would, unless the risk is very high.

Manoj Samanta said...

Evolution story you may enjoy -

http://phenomena.nationalgeographic.com/2014/01/06/evolution-hidden-in-plain-sight/

Ken Weiss said...

Oh, that's just bacteria! How could that have anything to do with, say stroke or cancer or athletic ability or IQ or.... in humans? No, what we need are bigger studies where we collect genomic data in ways that could hardly address such issues, and aren't pressing as hard with focused studies that might be more revealing!

Holly Dunsworth said...

On the phone with my mom last night she brought this belief up totally independently... that people want a pill to make it so they can stop worrying about their behavior so much. Particularly, they can stop worrying about what they eat if they just pop a few vitamins and supplements.. .which turn out to apparently be mostly snake oil.