testing

Of some interest with respect to DNA databases and privacy concerns: "DNA links 1991 killing to Colonial-era family".

The DNA sample was taken in the death of 16-year-old Sarah Yarborough, who was killed on her high school campus in Federal Way, Washington, in December 1991. The King County Sheriff's Office has circulated two composite sketches of a possible suspect -- a man in his 20s at the time with shoulder-length blonde or light brown hair -- but had been unable to put a name to the sketch.

In December, though, the department sent the DNA profile to California-based forensic consultant Colleen Fitzpatrick. Fitzpatrick compared the profile to others in genealogy databases and found the closest match was to the family of Robert Fuller, who settled in Salem, Massachusetts, in 1630 and had relatives who came over before him on the Mayflower.

This is a Y chromosome match based on the genealogical research of people who may be completely unknown to the "suspect". Fitzpatrick offers that a Y-chromosome match may be expected to share a surname, which is probative in the forensic situation. Obviously there are many possible scenarios in which such information will not lead to discovery of a suspect: the chance of non-acknowledged paternity events across 200 years is very high. I don't view the result as strongly actionable, but I do think it raises important questions about the future of genealogy databases.

We are near the time when whole-genome sequencing will make this kind of identification much more likely because unique genetic matches to 3rd and 4th degree relatives will be plausible. Finding a handful of rare mutations shared between a crime scene sample and an individual in a whole-gneome database would be a strong indication of a relationship. It's possible that the databases for whole genomes will grow faster than the technology will allow reliable whole-genome sequencing from a crime scene sample. So in this case, the issues with database use may be primary.

It would be an interesting exercise to estimate the fraction of unknown samples from crime scene Y chromosome and mtDNA that could be matched to a 10th-degree relative in the Genographic (or any other large) dataset.

G: Guess what Daddy and I learned last night? I'm more Neandertal than he is!

S: How did you find that out?

G: Our genes.

S: That's creepy.

G: What do you mean, creepy? We think it's awesome!

S: Awesome... in a creepy way.

Misha Angrist turns on the sarcasm filter for a proposal to discard raw data that may trouble research subjects ("If you want to destroy my sweater"):

Pay attention, kids: If it poses an ethical problem, then the obvious thing to do is to just throw it away! Delete it! Burn it! Shred it! Avert your eyes! The patient/research participant/taxpayer won’t mind! Trust me!

This is so annoying. It's cheaper in many contexts to do genome-wide genotyping than assay specific gene variants. So we'll increasingly see gene testing done on whole-genome platforms of various kinds.

But doctors don't order clinical tests for whole genomes, they order particular genetic tests. It's an obvious strategy for a testing company to provide only the ordered results, and either retain or discard further data, in the hopes of additional sales later. The company can upsell its "filtered" service as including additional validation or additional interpretive information of the kind that software can automatically add (for example, short-range phased haplotypes).

Angrist references a suggestion from an academic paper that a subject's APOE status should be blindly deleted from such results, to avoid the necessity of informing the subject about Alzheimer's risk.

This is the sort of thing we need to be thinking harder about -- how to alert unsuspecting people to minor or moderate risks that will be routine in whole-genome data.

The NY Times covers a new genome-wide association study of SNP variants and response to exercise ("Is Fitness All in the Genes?").

The phenotype is improvement in maximum oxygen consumption volume. Some people have rapid improvement with exercise and others don't. Straightforward enough, and there is one SNP that accounted for 6 percent of the phenotypic variation, which is quite strong as far as these associations go. Usually GWAS associations explain a much smaller fraction of the phenotypic variance.

The final few paragraphs of the article irritated me. It's like these stories have to follow a form, with a long disclaimer at the end. They report the facts -- variant explains 6 percent of variation -- and then they proceed to preach about how the facts may not matter:

“It will be years, if ever,” said Dr. Bouchard, before gene tests exist that can reliably separate high and low responders. Even if and when such tests become available, he continued, the results will not constitute an excuse for skipping workouts. “There are countless other benefits provided by exercise,” he said, apart from whether it raises your VO2 max. “Exercise can reduce blood pressure and improve lipid profiles,” he said. It can better your health, even if, by certain measures, it does not render you more aerobically fit.

More fundamentally, Dr. Bouchard said, elements of the interplay of genetics, environment, the human body and resolve probably always will remain mysterious and stubbornly individualized, no matter how much science disentangles the genome. People who don’t have an ideal version of the ACLS1 gene to prompt aerobic improvements from exercise, for instance, might harbor a different, unidentified gene that just makes exercise feel enjoyable, regardless. So, too, might someone whose body is genetically predisposed not to respond aerobically to running blossom during weight training sessions.

Hello? Six percent of the variance is six percent of the variance. The article ought to just say that 94 percent of the variance is not explained by this SNP. That answers the question! That unadorned fact tells you that the SNP isn't strongly predictive about exercise response for any single person. To do even better, the article ought to tell us how much of the total variance is explained by all the SNPs together.

I know, statistics can be difficult for NY Times readers, but honestly explaining the result would take a lot less space than what the article does do, which is to give us a long litany of "mysterious and stubbornly individualized." And the "different, unidentified gene that just makes exercise feel enjoyable, regardless."

Come on, people! Why not just tell us about earth spirits and auras?

The moralizing always goes the same direction. You'll never hear hand-wringing about how we can't trust exercise because it only predicts a small proportion of the overall variance in mortality risk. What about the "mysterious and stubbornly individualized" people who are healthy at 80 without ever lifting a barbell?