Software Glitch Scrambles Personal Genomics Data

By Kevin Davies

August 27, 2009 | In a commendably thorough piece of investigative journalism, New Scientist San Francisco bureau chief Peter Aldhous has uncovered an intermittent but potentially alarming glitch in the presentation of dozens of variants in his genotyping results from a major consumer genomics firm.

The data in question concerned the presentation in a web browser of several dozen single nucleotide polymorphisms (SNPs) in the mitochondrial DNA reported by deCODEme. The glitch, which Aldhous revealed to deCODE two weeks ago, did not impact any genetic conditions or medical information, and the source of the problem, Aldhous has been told, has been identified and corrected. But the episode reveals another layer of potential error and confusion in the consumer genomics world.

In the 20 months since the launch of deCODEme and 23andMe, firms that offer clients the opportunity to genotype 500,000-1 million SNPs, most of the debate about the value of the firms has centered on analytical and clinical utility. How accurate are the SNP calls on the various genotyping platforms? And how useful is that information in determining a person’s relative risk or medical outcome? Almost taken for granted has been the quality of the software and computing infrastructure used to deliver those data – a not surprising assumption, given the well-publicized relationship between 23andMe and Google and deCODE’s long-standing expertise in informatics and database management.

In meticulously comparing his 23andMe and deCODEme data, however, Aldhous recently made the surprising and serendipitous discovery that no fewer than 44 SNPs in his deCODEme mitochondrial DNA results, as viewed in deCODE’s state-of-the-art genome browser, were sometimes presented incorrectly. Aldhous was able to pinpoint the mistake to the graphical presentation rather than the actual genotypic data by comparing the raw SNP data downloaded from deCODEme and 23andMe, which use essentially the same Illumina platform. The raw data agreed entirely (except for two minor miscalls).

Aldhous said the erroneous presentation, which he describes in an opinion piece in New Scientist, was intermittent, with no apparent reason as to when his results were presented incorrectly. He could even view the mitochondrial data simultaneously on two computers – with one presentation correct, the other incorrect. (There was no problem encountered with the 23andMe presentation.) One genealogist, presented with the conundrum, even said: “Are you sure this is Homo sapiens?”

Aldhous relayed to Bio-IT World what he has learned from deCODE, namely that the error is due to an inconsistency between 32- and 64-bit environments and has now been fixed.


Click here to login and leave a comment.  


Add Comment

Text Only 2000 character limit

Page 1 of 1