Saturday, March 05, 2005

Ripping someone a new asshole using only a single genome.

You know what I love about science? Someone makes a crackpot claim, and everyone in the research community is there to jump on it and show how back-ass-wards the proposition is. Take, for example this paper that appeared in Nature last year in which the authors propose a new method for detecting natural selection using only a single genome.

Traditional methods for detecting natural selection using DNA sequence data require at least two sequences to be compared. For example, you may have the sequence of a gene responsible for nerve cell development from a human and the homologous gene from a chimpanzee. You hypothesize that evolution of this gene along the human lineage was responsible for some neurological innovation that lead to a cognitive difference between humans and chimps. In order to test your hypothesis, you must determine if natural selection acted on that gene in the human lineage. You could look for an excess number of amino acid substitutions between the two sequences, which would suggest rapid protein evolution along one of the lineages. This type of selection is known as positive selection. Conversely, a deficiency of amino acid changes would suggest purifying selection (selection for constrained function). Importantly, all accepted methods for detecting natural selection rely on comparing multiple sequences.

The new method introduces a metric known as codon volatility, based on the concept of mutational robustness. The authors propose that a gene under strong purifying selection will be robust to mutations that affect its function, and this robustness should be detectable based on the codons present in the gene. More volatile codons (those that are less robust to amino acid changes) should be common in genes under positive selection, whereas genes under purifying selection should have less volatile codons.

When it was first published last April this article raised some eyebrows in my department as many people were skeptical of the methods. Our doubts were confirmed as critiques came pouring in over the past few months from leading researchers at UC Davis, Cornell, Univ of Chicago, Nottingham, South Carolina, Univ of Houston, and Michigan. These articles were published in Nature, Molecular Biology and Evolution, and Genetics. Among the criticisms of the method are that it:
That's what you get when you try to support your claims with hand waving! I'm very surprised that this article made it into Nature as it doesn't even appear to be statistically rigorous enough for Genetics or Mol. Biol. Evol. They authors hold their ground in defense of their article, and only time will tell if their method will fall by the wayside or make it into mainstream molecular evolution research. At least they've got a step up on the IDers who haven't even published anything original.

For some reason, Nature loves publishing flashy, yet crappy, science. My favorite Nature faux pas comes from Dancing Naked in the Mind Field by Kary Mullis (the inventor of PCR). In his book he discusses submitting a letter to Nature about astrophysics as a graduate student. He admits that he knew very little about the subject, but it was published none-the-less. After coming up with PCR, he submitted his paper to Nature thinking that an idea that revolutionary would definitely be accepted -- it was not!


At 7:24 PM, Anonymous Anonymous said...

to my anonymous friend:

it seems that you seem to be in need of a second asshole yourself, since
you're so entirely full of shit. please allow me to rip you a new one
(as you so eloquently put it).

it is short-sighted, sorry excuses for (amateur?) scientists like yourself
that make evolution such easy prey for the creationists. simply because
a bunch of professors (hungry for publications but starving for any new
ideas of their own) write misguided criticisms of solid work does not make
that work any less true or accurate.

let's take your criticisms of volatility in turn. you claim that

> 1. Ignores effect of codon bias on volatility.

of course, volatility is simply a measure of codon bias- a measure that
happens to reflect selection pressures on genes. to say that the method
ignores the effect of codon bias on itself is nonsensical.

> 2. Is too sensitive to assumption of mutational process and unable to
> detect positive selection.

the only work to suggest a strong dependence of volatility on mutational
processes is by nielsen and hubisz. but as shown by plotkin et al (in
their response which appeared in nature), the simulations used by those
authors ignored population variability and were thus irrelevant as far as
volatility is concerned. those same authors also claimed volatility was
unable to detect positive selection, but this was based on the same set of
flawed simulations.

> 3. Only detects usage of a amino acids.

this is perhaps the least intelligent thing one could possibly say
regarding volatility. even the most rudimentary understanding of the
method makes it clear that the method controls precisely for amino acid
content. it is a mathematical impossibility for the amino acid content
to affect volatility p-values. this was of course made clear by the
original nature paper.

> 4. Makes erroneous assumptions regarding the distribution of synonymous
> substitutions.

this is again a hideous misconstrual of the method. there are no
assumptions regarding the distribution of synonymous substitutions made
by volatility.

> 5. Is not consistent with other, well established, methods for detecting
> selection.

volatility was not designed to replicate other "well established" methods.
it detects selection on different time scales and in a relative, rather
than absolute, sense, as was emphasized in the original nature paper.

>6. And positive selection has no effect on codon volatility in
>theoretical analyses.

this is the same as part of your criticism #2 above. rather than repeat
myself as you seem to enjoy doing, you may just read my response again.

for the true aficionados among you, may i suggest you read the follow-up
work by plotkin at all, posted at the arxiv preprint server. this work proves that volatility does in fact detect selection pressures, using detailed population genetic equations and simulations.

of course, i do not expect the fact that you are wrong to dissuade you
from your apparent biases against volatility. it is unfortunate, though,
that both you and the authors of the garbage you cite are unable to
separate fact from fiction. in that vein, it's rather amusing that the
mcdonald-kreitman test- now one of the most wldely used methods for
detecting selection from polymorphism and divergence data- received
similar negative attention in the literature soon after its publication,
with letters in nature from such luminaries as masatoshi nei, wen-hsiung
li, and dan graur (the last of whom is quite critical of volatility as
well) claiming that this heretical method could not possibly work.
perhaps the field of molecular evolution is simply so resistant to change
that any method smelling of originality is immediately ostracized, until a
few people with the capability of independent thought are brave enough to
support it. it will be interesting to see how long this acceptance
process will take for volatility (my guess is that it'll be a while).


Post a Comment

<< Home