Tuesday, August 23, 2011

Relying on the "experts"? An X-ample of the problem

There are many reasons why we must rely on expert advice in teaching, policy making, and understanding the world.  Too many aspects of modern life are just far too complicated for ordinary people to be sufficiently familiar with them.  Of course, experts are human and have their biases and limited knowledge, but we expect them to be forthcoming about areas of doubt, and to know at least the main outlines of their field.

When they don't, the arena for manipulating facts to serve vested interests is widened relative to what it should be.  Then policy is in limbo, is hijacked by those interests, or becomes essentially uninformed--the exact opposite of what is needed and expected.  The problem applies to scientists even within their own presumed areas of expertise.

Here's an example of this, from genetics, our own field.  The current issue of the prestigious (expert-driven?) American Journal of Human Genetics has a picture of 16 US 25-cent coins on the cover, some heads up and others tails up.  The picture is meant to illustrate a story about X-inactivation.

The dogma has been a phenomenon known as Lyonization, or X-chromosome inactivation: early in a female embryo, each cell randomly selects one of its two X chromosomes, and wraps it up tightly (so to speak) so that its genes cannot be activated.  Every descendant cell in the later embryo and through the individual's life is assumed to 'remember' which of its X's to activate and which to keep inactivated, so that an adult female is a mosaic--some patches in her tissues are expressing the variants of the X that she inherited from her father, while the other patches express those inherited from the X donated by her mother.

This means each female cell has one active X, and the idea was that this keeps gene dosage from the X in her cells compatible with the rest of the genome, since males only have one X. Since X-inactivation is random, roughly half a female's cells have paternal, and half maternal X activated.  Because X-inactivation occurs early and randomly, every female is different in terms of which patches of her cells are which.

Scientists needing melodrama rather than some dull picture of, say, actual chromosomes, the AJHG decided to illustrate probability with a coin.  The imaged theory is 8 Heads, 8 Tails.  But the cover story makes the point that evidence now shows that the two X's don't necessarily have an exactly 50-50 chance of being activated in a given cell.  So, if genes on one of the X's give a growth advantage, more of the female's body may be found to be expressing that one.  Similarly, if one X has damaging genes, fewer than half her cells will be found using that one.

This is interesting, and not surprising, even if it hadn't been documented as well in the past.  But what is surprising is that it is (or should be) by now well-known that Lyonization does not really work like this!  Indeed, various people including the lab of Dr Kateryna Makova here at Penn State, have shown that only part--around 85%--of the X is subject to inactivation.  In female cells both X's remain active for the other 15% (raising questions about the dosage-balance view in regard to the genes in those regions).  Indeed, there may be variation, even probabilistic variation from person to person or cell to cell, in which parts of the  X remain active, and one possibility is that this has to do with various bits of transposable (move-around-able) DNA that may have been inserted there over the ages past (a subject beyond this post).  Makova and colleagues have shown that the active regions seem to have been preserved by natural selection from varying as much as the regions subject to inactivation, presumably reflecting something about function, but beyond our point here.

This incomplete inactivation is entirely different from what the AJHG cover story is about.  But it's a leading journal, and it is publishing what it basically portrays as a surprising deviation from classical theory....without even acknowledging that the theory has already had a major revision well-known and prominently published.

This may be a minor point in the scope of science (but perhaps not in understanding chromosome and sex evolution!), but it shows how subtle can be the intercalation of misunderstanding (if not dogma) in science. We're not talking here about journalists' mistakes (except to the extent that, in the huge arena of major journals, editorial staff looking for cool cover pictures are people with journalism degrees rather than scientists).  The AJHG is one of the major journals in a highly sophisticated field, so one can imagine what the proliferating commercially driven hasty journals are like when it comes to reliability and accuracy.

We all are vulnerable to mistakes, so we're not particularly picking on this one, beyond making the point about the grip of dogma, the reliance on theory once learned but hard to relax, and the problem of anyone, even scientists, even in their own field, and teachers at all level, keeping up to date. It shows the extent to which even science is a cultural phenomenon, that involves lore as well as 'fact', but upon which opinions are framed and policy is made.

And speaking of imagery and dogma, by the way, why don't you try something:  take a handful of quarters, and flip them vigorously a number of times.  Do they actually come up half heads and half tails?  In fact, if you do enough flips, you would find that they don't.

10 comments:

Alan Packer said...

Actually, the AJHG cover shows 13 heads and 3 tails (not 8 of each). Also, the cover story (if you read the accompanying blurb carefully) is not referring to new research published in the current issue of the journal. It's referring to work over the last 30 years that has indeed established the idea of skewed X-inactivation. The AJHG these days frequently uses cover images that illustrate an important point in the development of genetics in the past, rather than focusing on newly published research. The editors there got it right.

Ken Weiss said...

Well, yes, the cover blurb was clear about the point being presented, and that's why the 3/13 distribution rather than 8/8. I think our blog post was clear that the point of the cover story was as you say, that evidence suggests that 8/8 is not what one sees and we know some reasons why.

But to me, at least, it perpetuated a seriously outdated view of X-inactivation, that it is not complete within the X. The image, as we tried, perhaps not clearly enough, to say was misleading in perpetuating the idea that Lyonization is complete.

It's a separate point from internally incomplete X-inactivation. To me, rather than make it seem as if the 'new' finding is that mosaicism is not exactly 50-50, which is interesting and we have no reason at all to question it. Indeed, meiotic drive in germline transmission and somatic mono-allelic expression across the genome are related to the X phenomenon.

The point we were making, and I think it's legitimate, is that the cover story perpetuated a dated model of X-inactivation. Had the blurb said something about the fact that the complete chromosome itself was not completely or uniformly made hemizygous even within individual paternally, or maternally-inactivated cells, it would have been more acceptable, at least in my opinion.

I would feel the same even if this were in the NY Times of BBC science news, but perhaps less strongly than what's in a major journal.

The deeper point is how very difficult it is, even for professionals to keep current in their own field these days. Yet much in terms of research, and especially perhaps in terms of public commitments to one vs another area of research, depends on science doing so, I personally think.

Alan Packer said...

Here's the first sentence of the blurb: "In females, one copy of the X chromosome is inactivated so that most X-linked genes are expressed at the same level as they are in males". I think we can infer from the use of the word "most" that editors Williamson and Bungartz (both of whom have Ph.D.s in genetics) are well aware that some genes on the inactive X escape inactivation. That they chose to highlight a different point about X inactivation, without exhaustively reviewing everything we've learned about the subject in the last quarter century, seems to me entirely fair.

Ken Weiss said...

I see your point. To me, this is less of a surprise than the partial inactivation and the mechanisms, and the way the blurb was written to me perpetuates a dated view of X-inactivation. That the writers have PhD's is not very persuasive since these areas are so huge that nobody can remotely keep up with all of them, and the challenge is to know a given area well enough to comment in an up to date way. I don't by the way know if their 'most' referred to the known recombining area between X and Y.

Anyway, we've made our respective points and that's a good aspect of blogs--that things can be said and responded to....

Alan Packer said...

Thanks for your reply. I actually completely agree with you that it's become near impossible for even well informed scientists to stay current.

Ken Weiss said...

and it's serious, because of the butterfly effect. Being slightly out of date can lead to research being launched that gets off on a tangent that can flow on its own and get things even farther away from a more efficient or effective understanding.

Nate Davis said...

"take a handful of quarters, and flip them vigorously a number of times. Do they actually come up half heads and half tails? In fact, if you do enough flips, you would find that they don't."

Doesn't the law of large numbers dictate that as flips(n) approaches infinity, the proportion of heads should converge on 1/2? You know, with ENOUGH enough flips. Or am I missing some deeper point? :)

Ken Weiss said...

It only converges on 1/2 if the coin is 'fair'. If you roll dice, at least the cheap ones that come with most board games, in fact they don't come up exactly 1/6 for each possible side.

If a coin is not 'fair' then one side will come up a small fraction of the time more than one over large size trials. If the probability (that is, based on the structure of the coin) is, say 0.55 Tails rather than 0.50, you will converge on 55% Tails.

The issues are subtle including how to tell if the flips themselves are fair and so on. In fact, there is even the question about whether the end result is random or not--that is, is the variation due to something inherent in the coin, or is it due to aspects of the flipping?

The answer to that question is known.

Nate Davis said...

So what we need is a machine that will perfectly flip within strict tolerances, and either perfect coins or (less ideally) a separate coin for every flip event...

I see your point, yes, and its broader implications.

Ken Weiss said...

Such a machine has been built, once recently and I believe once in the late 19th century. The coin comes up the same way on every flip. The process itself is essentially deterministic and the apparent randomness when you try it yourself is due to flipping variation.