Genome Hacker Uncovers 13-Million-Member Family Tree 61
ananyo writes "Using data pulled from online genealogy sites, a renowned 'genome hacker' has constructed what is likely the biggest family tree ever assembled. The researcher and his team now plan to use the data — including a single uber-pedigree comprising 13 million individuals, which stretches back to the 15th century — to analyze the inheritance of complex genetic traits, such as longevity and facial features. In addition to providing the invitation list to what would be the world's largest family reunion, the work presented by computational biologist Yaniv Erlich at the American Society of Human Genetics annual meeting in Boston could provide a new tool for understanding the extent to which genes contribute to certain traits. The pedigrees have been made available to other researchers, but Erlich and his team at the Whitehead Institute in Cambridge, Massachusetts, have stripped the names from the data to protect privacy."
Source material is unreliable (Score:5, Insightful)
The online family tress they're using as source material are notoriously unreliable. They don't include sources, errors are copied from tree to tree by name collectors, and many links are often incorrect. I can't believe they think they can draw a conclusion from any of it. Respectable genealogists would laugh at this endeavor.
everyone's a hacker (Score:5, Insightful)
Wow, now everybody's a hacker these days. It started to go downhill with the whole "lifehacker" meme. perhaps I should be called "garbage hacker" instead of the prior preferred term, "sanitation engineer"
Re:Source material is unreliable (Score:5, Insightful)
Even if all the information were a 100% accurate representation of the actual records and all links were correct, the original records likely contain numerous errors or important omissions; to take the most obvious point, there is likely to be almost no way to verify whether children were legitimate or not. So its usefulness for genetic study seems doubtful to me as many generations later I suspect those sort of effects are difficult to pick up or isolate properly in living people's genes.
What's worse, in some historical periods it would not have been uncommon for some children to be biologically unrelated to either of their legal parents - e.g. lovechild of an affair the man had with a woman who was also sleeping with other men (but who claimed he was the father as he represented the best economic/social prospect of the possibilities), after which the man might take responsibility and raise the child as his own.
Re:Source material is unreliable (Score:4, Insightful)
The article does comment on this.
If you're using a maximum likelihood analysis, your model can allow for unreliable data. E.g. you could assign a 10% chance that the paternity is not as recorded. Then you would have probability calculations like
P(child inherited gene from father)=0.9*P('father' (according to genealogy) had the gene)+0.1*P(random male in the population had the gene).
You can even make the 'false paternity rate' a parameter in your model, so the data itself will tell you what value is best. However, if the data is too unreliable, all that your maximum likelihood analysis will tell you is "we can't conclude anything from this data". (Assuming you correctly model the unreliability. If you don't, your analysis is liable to give false results.)
Maximum likelihood is not always computationally feasible, depending on the model you're trying to fit and how much data you have.