I thought I’d also mention that one problem with data mining is a lot of the information can be crap and/or mis-annotated.
I think I saw a stat once that said something like 30% of the sequences in GenBank were incorrectly identified.
I took a bioinformatics course once and one assignment was to identify a DNA sequence. Simple enough–just BLAST it. Let’s say it came up as matching the human Fluxomioma gene. The trick of the assignment was to use other tools and correctly identify the sequence as being primarily a DNA vector sequence–contamination–that some idiot had submitted and called human Fluxomioma gene.
That’s why the better databases, and probably all of the commercial proprietary databases, are manually curated in some fashion. It’s a big selling point.
To use Company X as an example, just because I’ve used some of their software:
I was also thinking about doing this. But I’m concerned about companies (especially US companies) having access to my DNA and with that hackers/governments/criminals.
What if Skynet* ever comes into power and wants to clean the world. Should I be worried?
[hijack]
Oh god yes, the amount of shit out there is mind-boggling. I once helped someone troubleshoot an RNA-seq experiment which had 90% of the sequences that didn’t map to the correct organism’s genome. After talking in circles for a while, I just told her to BLAST a bunch of the non-mapping reads. All of these reads matched genomic sequences from a truly random grab bag of organisms. I recall some sort of marine parasite, an endangered species of tree, and about a dozen other organisms.
After much more troubleshooting, it turns out that these sequences were added by the library preparation kit. Now these were NOT the Illumina adapters or barcodes, but another set of proprietary adapters that were added by the kit and supposed to be removed during the library preparation. Apparently these adapter sequences can concatamerize and completely take over the library. The company that made the kit never disclosed the sequence of their adapters, which made it impossible to completely remove them from the sequencing results.
But several research groups never even bothered to figure this out, and published genomic sequences that are almost entirely comprised of concatamers of proprietary DNA sequences. It’s not just isolated contigs, either: since some of the adapter sequences are attached to real reads, there are giant chunks of concatamer shit in the middle of otherwise plausible looking genes.
Sometimes I feel like giving a biologist access to a next-gen sequencer is akin to handing loading guns to a preschool class.
[/hijack]
Let’s say 23andme identifies a rare genetic variant, and tells you that it’s associated with Fluxomioma. If you dig, it could easily turn out that the only study demonstrating this association was published by the University of East Elbonia, West Elbonia Campus. And their sample consists of twenty people with Fluxomioma in the nearest hospital, while the control group was twenty friends and family of the graduate students tasked with getting samples.
For every variant that 23andme reports, you have to trust that (1) their technology correctly identified it, (2) 23andme gives you an accurate summary of the scientific literature on that variant, (3) the literature includes reliable studies, and (4) that the studies can be generalized to your circumstance.
That gene variant may actually cause Fluxomioma only in a tiny isolated village in East-West Elbonia where the population has remained isolated for centuries.
Sorta related. I once worked with someone who had a favorite gene. It hadn’t been found conclusively in the model organism (another class of vertebrate) she was working with, but had in mammals. One (and only one) group had a paper saying they had found this gene and another group, who actually was involved in the genomic sequencing, had trashed it pretty convincingly.
The sequence the first group had reported was like 95% identical to human, while the similarity between human and mouse (and rat) was like 80%. It was unlikely that the gene would be more similar between organisms in different classes than within.
However, the first group submitted their sequence and so she found it in a search.
“This proves that the organism has the gene!!!” she triumphantly exclaimed. “I am right! You all now look foolish for doubting this/me!”
I asked her what her search results would be if someone took that sequence, changed a few nucleotides, changed the species name, and submitted it. Would that prove the the new species had the gene? Because, according to her criteria, it would.
Any hints for me on who made the kit with the concatemers?
Eh, what the hell. The group with this problem published some of their efforts to clean up the concatamer mess and a quick google reveals other people discussing the same problem. It was a Clontech SMART cDNA synthesis kit (not the library prep kit, I mis-remembered). Therefore the spurious BLAST hits must have been transcriptome rather than genome sequences.
I had it done out of curiosity and for fun. I’m not so daft to think because I got a note about some genetic disorder that I’d take it as gospel. I’d go to a doctor and have it checked out.
I haven’t done as much as I should with it. I loved finding out my ancestry (a fraction Eastern European Jewish, East Asian - as a US born Black dude didn’t see that coming) and a few distant relatives that are way more into this than I am.
Meh. They do a pretty good job of explaining that there is a fair amount of uncertainty on many items and of grading the degree of confidence they have in the research reports that base their disease risk descriptions. The overstep seems mild to me. The believing that they did not need to follow all the rules (that are in place for good reasons) was hubris likely a result of the Google association. They’ll recover.
Of the two I would only be interested in the National Geographic one, I am more interested in where we have genetic input from. I already knows that parkinsons pretty much runs rampant on one side of the family, and alzheimers runs on the other. 2 of 3 brothers [father and one uncle] had wet macular degeneration. So I am pretty hosed medically speaking. [I already have neurological issues starting but they are very minor and being watched.]
We do actually have roughly 400 years of geneology on my Mom’s side, and roughly 800 on my Dad’s side [one vine has been traced in church and tax documents and grave markers back to the mid 1100s in northern France only because that vine was lesser nobility and kept records for inheritance purposes. Where there us a will there is a greedy relative :smack: Though it was interesting seeing the ping ponging back and forth across the borders of what are now the Netherlands, Belgium, Flanders and the Nord-Pas de Calais region.]
Yeah, this occurred to me too, since in one sense they were literally in bed together.
Being disregarded also seems to have the FDA’s bureaucratic dander up. The letter read like “You’ve been blowing us off so maybe this will get your attention”.
I agree. I work with statistics and probability a lot in my business. My experience with MDs and those in the gene analysis biz is that they wouldn’t know a statistically valid piece of analysis if it came up and bit them.
Well, to be fair, 23andme does look to hire statisticians. I posted this job description in Post #39:
On the other hand, I do agree with you that a lot of people doing research don’t have the first clue about statistics.
I worked in a lab where the PI was a MD, and most of the postdocs were MDs–so they actually had never had a lab position or done bench-top research prior to their postdocs. They had about as much experience with statistics as they did with pipetting. We had a lab meeting once where the PI asked what a p-value was and I was the only person out 12 of people at the table who could correctly tell him (somebody actually said it proved your experiment worked technically, if it was less than 0.05).
This lab once published a paper where they reported their experimental replication as “n=a pool of 50 samples”.
For my friends, I used the analogy of "You’ve seen those commercials where the actor says “I’m not a doctor, but I play one on TV’. These people are like ‘I’m not a scientist, but I play one at work’.”
So, I think the FDA does have a legitimate concern about the general public (or even MDs) trying to interpret “health recommendations” based on the genotyping in these reports.
I think the Illumina SNP platform they use for genotyping is reliable, it’s just not approved for this purpose.
