deCODEme's embarrassing data processing glitches - lessons for companies and customers

Late last week I noted an intriguing offer by personal genomics company deCODEme: customers of rival genome scan provider 23andMe can now upload and analyse their 23andMe data through the deCODEme pipeline. 
On the face of it that's a fairly surprising offer. As I noted in my previous post, interpretation is what generates the real value for personal genomics companies, so giving it away for free seems a bizarre approach to business - especially for a company living on the edge of a financial precipice. However, I also argued that the intention here is likely to be to generate an opportunity for deCODEme to display its interpretation skills to otherwise entrenched 23andMe customers, in preparation for the upcoming battle for interpretation supremacy in the whole-genome sequencing era.
In digging back through my archives I realised that this isn't actually the first time that this strategy has been employed in the personal genomics game: back in June this year, 23andMe offered its interpretation service free to customers of Illumina's freshly-launched $48,000 whole genome sequencing service (the original source is this subscription-only article in industry publication In Sequence).
It's nonetheless the first time that a personal genomics company has opened itself up to genome scan customers, and it's certainly a disruptive (and potentially game-changing) move. AccessDNA's Jordanna Joaquina even goes as far as to speculate that this may herald a shift in deCODEme's strategy towards pure data interpretation. I personally think this is unlikely for deCODEme itself, but I wouldn't be shocked to see a proliferation of multi-platform interpretation services over the next 12 months (Knome's recently announced discovery service is a step in that direction).
But creating an interpretation service that can deal seamlessly with data provided in a multitude of formats from different providers can be a challenging task, as deCODEme learnt in a particularly embarrassing manner this week:
Nothing seems to be going right for DeCode Genetics, the Icelandic personal genomics firm that declared bankruptcy last month. In an attempt to win customers from its Californian rival, 23andMe, DeCode has offered free analyses for anyone who uploads their 23andMe data onto its website - only to produce mangled interpretations for genetic ancestry.

More alarming, from my point of view, the company also suggested that I have a 40 per cent lifetime risk of developing Alzheimer's disease - which if didn't know better thanks to previous scrutiny of my own DNA, would have been a devastating thing to learn.

The article, from New Scientist's Peter Aldhous, details a series of blunders in deCODEme's management of the 23andMe data transfer, compounded by an embarrassing "Reply All" incident in which deCODE's press officer Edward Farmer accidentally forwarded a dismissive email back to Aldhous.
The ancestry problems (outlined in detail by David E.) seem to have resulted from deCODEme relying on an outdated data-set from 23andMe to test their software, something that was rapidly fixed (see screenshot from my own data below). But the Alzheimer's risk prediction error is the result of using an imperfect proxy marker within the APOE gene - a seriously dangerous thing to do for a variant with an unusually large effect on disease risk.
What's the big picture here? Well, obviously 23andMe customers porting their data over to deCODEme should take the results with a grain of salt - but that's true for anyone looking at personal genomic data anyway. These errors shouldn't discourage anyone from checking out their data on deCODEme's interface, and indeed I'd strongly advocate that all 23andMe customers do so in order to get a second (imperfect, but nonetheless informative) perspective on their own genetic data.
This is also a timely warning to everyone - both customers and companies - starting to move through the personal genomics space. The next few years will see a proliferation of data formats and interpretation services, and there will inevitably be errors made due to clashes like this one. It's important that companies do their best to minimise the effects of such errors (and deCODEme could certainly have done a better job here); but it's also crucial that personal genomics customers follow Aldhous' example in constantly questioning their data, finding inconsistencies, and following them up until an explanation is found
This is your genetic data, and you deserve an accurate interpretation - but in these early, chaotic days of the personal genomic industry, that means playing an active role in interpretation yourself.
My ancestry data immediately following my initial upload on 17/12/09 (note the bizarre X chromosome results):i-d1f9ca37b5e9b40f8e213117796da6df-chrom_painting_091217.jpg
And then following the bug fix, today:

More like this

Clearly what is needed is a standardized format. If deCodeMe is really going to push this business model (someone else generates, we analyze), then they should be pushing for a standard. But in any case, the academic world should be hammering one down (a) so that it is neutral and not biased for or against any platform or party and (b) because generating new genetic interpreters (ala SNP-o-matic) will probably be a bioinformatics ore seam for quite a few years hence.

I wouldn't be surprised if 23 did this intentionally to mess their shiznit up. Nelson laugh, ha ha, the bankrupt and about-to-be-bankrupt company messing with each other.

Here is my prediction: Risk of Death 100% with or without Alzheimer's.

I am a 23andMe customer and have posted my data to deCODEme. I take both companies predictions as predictions, nothing more. Not quite crystal ball gazing but not far off. Those SNPs tested are not genes, are mostly repetitious nucleotides of no meaning, junk pure and simple. They have been found to indicate the way certain genes function or malfunction, an association, but no where near 100%. I have the SNP which predicts brown eyes 80%, well I do have brown eyes but obviously there are people with the same alleles to that SNP who have other eye colors. Both companies can't tell you what your ABO blood group happens to be.

I sound disappointed. Not really. I have found insights into my heritage, a person related to me via Relative Finder, and SNP facts that I did not know and that surprised me but later things started to fit into place.

deCODEme is currently experiencing a glitch with their ancestry PCA eigenvector maps of one's place in the world.