PGP sequence data disappointing

The promise of release of raw sequence data files from the first 10 Personal Genome Project volunteers certainly caused a media stir (see the round-up by the PGP's own Jason Bobe), but the actual released data are pretty underwhelming.

So far raw sequence data files have been posted on the PGP profile sites of only four of the ten volunteers: George Church, John Halamka, Esther Dyson and James Sherley. The files are the result of targeted resequencing of a proportion (perhaps 20%) of the protein-coding regions of the genome (called exons, collectively the exome). Although a relatively small proportion of the genome as a whole, the exome is highly enriched for functionally important variation, so this small slice of sequence could actually be quite informative about the genetic variants associated with diseases and physical variation.

However, when I downloaded and examined the files my hopes weren't too high - participant Misha Angrist had already warned on his blog that the data release was not the world-changing event that the media might lead you to believe:

I have to say, this whole extravaganza felt more like a walk-through or a dress rehearsal. Several of us did not get our sequence data yesterday and those who did got very rough, low-coverage data.

After checking over the files I agree with Misha about the "rough, low-coverage" bit - the data released so far makes for pretty desultory reading. For instance, here's a snippet from Esther Dyson's file:


(For those who know and care about these things: the sequences are in FASTQ format, and I've stripped out the '+' and quality score lines for clarity.)

You'll note that the file is divided into short snippets of DNA sequence separated by headers (the lines preceded by '@'). There's about 55,000 of these fragments in the files, but a fairly hefty proportion are low-quality sequences with no real genetic information (the 'nnnnn' fragments above).

The files are difficult to interpret at first glance, with no identifiers indicating what each sequence entry represents. However on closer inspection each entry in the data files appears to represent a single sequenced exon covered by multiple reads from a next-generation sequencing platform. It's difficult to estimate the true coverage of these exons from the data provided, but the overall sequence quality certainly doesn't look great - here's a breakdown for the data from John Halamka as an example:

  1. 55,054 exons sequenced in total, with an average length of ~163 bases;
  2. overall, more than half of the targeted bases (56%) have no available sequence data (i.e. are marked 'n');
  3. 16,644 (30.2%) of the exons contain absolutely no interpretable sequence (i.e. are entirely 'n');
  4. of the remainder, 15,549 (28.2% of the total) contain 30 or more contiguous 'n's.

In total, these data provide actual sequence information for just 0.13% or so of the total genome - but even that fraction is likely to be a serious over-estimate, since many of the bases that have been called will have low coverage and thus be unreliable. Determining the reliability of each base call using the provided quality scores is not straightforward, but I'll be doing my best to sort this out over the next few days.

To be fair, the files are clearly marked "Preliminary exon data", and it's still very early days for the PGP - I would expect to see a dramatic increase in the volume and quality of the sequence data released over the coming months. However, given the hype surrounding this data release I'm a little disappointed by the data itself. Can genome sequence data really be said to be publicly available when they're dumped on the web in a flat text file without any gene annotation or explanation regarding its format, making it useless for anyone other than bioinformaticians?

We're in a crucial window of time here: the PGP (along with high-profile companies like 23andMe) is blazing the trail for the whole field of personal genomics, and the world is watching. I hope that future data releases from the PGP can give the world something to genuinely get excited about.

Subscribe to Genetic Future.

More like this

Well, c'mon Daniel. Sequence data isn't worth anything without the bioinformatics. ;-)

I've also mentioned this on

I've written a tool to recognize rs# sequences in the fasta, but it needs assembled fasta, not these short reads. I may try to do my own assembly, but at the moment time is too tight, and I expect it won't be all that cheap even with the amazon cloud.

What I'm disappointed by is the lack of the microarray data release. That has been completed since January 2008 as evidenced by

I'd like to prepare promethease reports for all of the PGP10, but none of that data has been released publicly except for 2 of them which were given to me directly.

Hi Daniel -

Thanks for the critical eye about the data. We're in the process of posting more information about the PGP data quality goals, how the preliminary data stacks-up, why we felt it was necessary and important to release preliminary data, and QC methods (e.g. concordancy across data sets, samples, etc).

Cariaso: We'll be posting genotyping data soon too.

Jason Bobe

Hi Jason,

Sounds good - but it would have been nice to emphasise how preliminary the data were before they were released publicly. I know it's hard to control what the media decide to report or ignore, but posting a definitive statement on the PGP website about exactly what was being sequenced, using what method and at what coverage would have been extremely helpful.

According to the UCSC genome browser, read number 228 in the post contains a 100% identical match to a coding exon for gene MCM10 on chromosome 10. But I agree that overall this preliminary level of data quality is not sufficient for any scientific inferencing.