Friday, January 29, 2010

The human touch

As mentioned in my previous post, I interviewed Emmeline Hill, PhD, geneticist at University College Dublin and co-founder of the new equine genetic testing company Equinome last week for Thoroughbred Times. You'll have to read the interview in the February 13 issue to get that copyrighted information, but, suffice to say, Dr. Hill is personable, straightforward, and highly confident that what she has discovered is valuable information for Thoroughbred owners and breeders.

I'm not so sure. One of the questions I asked Dr. Hill was whether her test is superior to the opinion of an experienced, competent conformation and pedigree professional. Again, you'll have to read the magazine to see her answer.

One of the reasons I have doubts about Dr. Hill's research is sample size. Hill's basic sample size was 148 elite Thoroughbreds. That's pretty small for a study of a population as large as the Thoroughbred. There are certainly more than 500,000 Thoroughbreds alive in the world right now. A few years ago, the number was pushing 1-million, but that has probably declined significantly.

Hill's elite sample gets divided up further into CCs, CTs, and TTs, naturally, producing even smaller samples of horses with those genotypes. One result of that small sample size can be seen in the standard error of the average best racing distance of the three groups.
Standard error is a measure of uncertainty. I'm sure you've all seen political polls that predict that candidate A will win 51% of the vote and candidate B 49% +- 2%. That “+- 2%” is the standard error. In that case, that means that the race is actually a statistical dead heat, since the real percentages for each candidate theoretically could vary by two percentage points and they're only two percentage points apart in the poll. Another poll the next day could (and often does) give exactly the opposite result.

Hill's results show that the best racing distance for horses with the CC genotype that her “speed gene te average st” tests for is 6.2 furlongs, +- 0.8 furlongs. That's not bad. That means that it's pretty likely that the average best winning distance for those horses is going to be between 5.4 and 7 furlongs roughly 70% of the time (in a normal distribution), and 6.2 furlongs is the most likely number. You're probably not going to do much better than that with such a small sample size.

The problem appears with the CTs and TTs. The average best winning distance for CTs is 9.1 furlongs, +- 2.4 furlongs; average for TTs is 10.5 furlongs, +- 2.7 furlongs. I'm not an expert—it's been almost 40 years since my graduate level statistics courses—but those look like pretty big standard errors to me....just the kind you might get from a human expert.

I am quite confident that with the appropriate pedigree information and a good look at the physical horse that I could predict its best racing distance within about a quarter mile pretty damned consistently.

Still, people tend to want what they perceive as certainty. I strongly suspect that many of the large farms and racing stables in England and Ireland have already opted to have all of their horses tested. To a billionaire, the $1400 per horse cost of the test is pretty meaningless.

I also strongly suspect, however, that the test will have virtually no effect on employment of bloodstock agents and other advisers. Equinome's test cannot tell you if that yearling with the genotype you prefer has offset knees or a curby hock, or whether he moves like a racehorse.

That requires a human eye, human intelligence, and human experience.

P.S. It may well be true that Dr. Hill has since gathered more unpublished evidence that reduces the standard error. But the spreads in the data actually make sense in terms of the way horses actually race. There are horses that are purely sprinters. There is another group that can win at sprint distances but are better up to about 9 furlongs. Then there is another group that generally can't beat decent horses at six furlongs but can win between 7 or 8 and 12 furlongs or more. And then there are horses that don't really fit within any of those patterns but are superior at every distance. Those are rare animals indeed.

8 comments:

  1. John,

    This is an insightful take on the issue that Hill has developed with her company. And as your conclusion states so well, it is nothing that will surpass human evaluation. Luck to us all.

    Especially the horses.

    Frank

    PS Do you recall the sample size Hill drew on in her work on the MtDNA traces back to the foundation mares?

    ReplyDelete
  2. John,

    Interestingly today another scientist, this one in Australia announced a slightly different path.

    Associate Professor Allan Davie has combined with scientists in China to study "Polymorphisms of mitochondrial DNA in relation to racing performance in thoroughbred horses". They just got a research grant from the Australian and Chinese Goverments to do the work.

    Byron

    ReplyDelete
  3. Sample size for that study was listed as 100, but sample size in that case is not so important because that study does not look at averages. It simply investigates what particular haplotypes the members of the group have. Each horse simply has a given haplotype or another, so it's not a matter of averages.

    ReplyDelete
  4. Thanks for that information Byron. Steven Harrison in Britain has published work relating mtDNA variations to winning distance over about 30 years worth of English races. The study was published in the journal Mitochondrion in 2006 and there was a review in Pacemaker...http://thoroughbredgenetics.com/PACEMAKER%20BG.PDF

    ReplyDelete
  5. I added a P.S. above, and am adding it as a comment as well....

    P.S. It may well be true that Dr. Hill has since gathered more unpublished evidence that reduces the standard error. But the spreads in the data actually make sense in terms of the way horses actually race. There are horses that are purely sprinters. There is another group that can win at sprint distances but are better up to about 9 furlongs. Then there is another group that generally can't beat decent horses at six furlongs but can win between 7 or 8 and 12 furlongs or more. And then there are horses that don't really fit within any of those patterns but are superior at every distance. Those are rare animals indeed.

    ReplyDelete
  6. Speaking of sample size, what is your opinion of the quite popular nicking grades that often are based on extremely small samples.

    ReplyDelete
  7. I don't think - and I certainly hope - that nothing can replace the human eye.

    Just looking at these articles gave me terrible memories of high school science class.

    If my business became an experiment in theoretical biology, I wouldn't be very good at it. I will stick with breeding the best to the best and hoping for the best, and leave the science to the scientists!

    ReplyDelete
  8. The end result of all this study into equine genetics could be some degree of genetic engineering- but only if we allow AI. That could mean that the boffins in the labs, funded by the wealthy breeders/owners, would be able to tinker before depositing the genetically enhanced embryo into the womb of a unsuspecting mother.
    Maybe this would produce the ultimate racehorse(s) but what about us little guys. Who wants to watch a collection of Sheiks and billionaires win everything.
    Some things are more about the competition than winning everything.

    ReplyDelete