PGP participants at the 2013 GET Conference. Photo credit: PersonalGenomes.org, license CC-BY

Around 150 Personal Genome Project (PGP) participants attended the 2013 GET Conference and it has become a tradition to take a group photo. Here is the group photo from last year’s GET Conference.

From our guidelines for PGP implementations:
“Non-anonymous. The risks of participant re-identification are addressed up front, as an integral part of the consent and enrollment process; neither anonymity nor confidentiality of participant identities or their data are promised to research participants.”

We have designed a consent process that includes many layers of upfront and ongoing education about the unique nature of public genomics research studies like the PGP. One of the important messages to participants is that their data are highly identifiable and therefore not “anonymous”. For example, the study guide that accompanies our mandatory entrance exam provides one of the more famous examples of how only a few pieces of demographic data can reveal a person’s identity:

From the PGP Study Guide:
“Identities can be discovered with surprisingly little information — for example, the combination of sex, birth date and ZIP code is specific enough to be uniquely identifying information for 87% of people!”

We know that hands-on demonstrations of otherwise abstract concepts can be extremely valuable for learning. Talking about a “personal genome” in the abstract can be a far different experience compared to wading through millions of variants contained in your very own personal genome sequence! So to enhance understanding of identifiability, we invited two research groups to demonstrate how re-identification is possible using public PGP data during GET Labs in Boston (April 25-26).

Latanya Sweeney’s Data Privacy Lab drew upon her pioneering work on the identifiability of demographic data to show how these techniques can be applied to public PGP profiles containing sex, birth date, and ZIP code. It was no surprise to find that many PGP participants are, in fact, identifiable. Indeed, all PGP participants should expect this potential outcome.

This is important, considering Harvard PGP participants are able to add ZIP codes to their public profiles in anticipation of research activities that explore how geographic location — and all the associated chemical exposures, microbes, viruses, air quality, allergens, etc. — impacts health.  For anyone who was not at the GET Conference, Sweeney’s group has created a tool showing how identifiable you are in your own zip code.  Check it out here: http://aboutmyinfo.org/

A word of caution is required here about the best way for PGP participants to respond: we strongly advise any participant concerned about the identifiability of their data to reconsider their participation in the Personal Genome Project. Another viewpoint, one that we find worrisome, is for participants in the PGP to deploy clever tricks for reducing the identifiability of their public data. As part of their demonstration, the Data Privacy Lab is providing tools to participants that “scrub” their data (e.g. replacing a 5 digit zip code with a 3 digit zip code, etc). This may create the impression of privacy, but it will not make participants anonymous. Earlier this year an exciting study published by Gymrek et al. in Yaniv Erlich’s lab forcefully demonstrated that genome data alone is extremely identifying. Melissa Gymrek also had a table at the GET Conference this year where she demonstrated the technique to participants. Their research matched whole genome Y-chromosome data to ancestry databases, which link surnames with Y-chromosome markers. With these surname clues and just a few other pieces of publicly available data, their group was able to identify specific individuals and families from their highly distributed “anonymous” cell lines.

Thus, all participants should believe that they are identifiable: there is no such thing as an “anonymous” genome!

In our experience, many participants want to be identified and are very open about which public profile is theirs. The PGP does not require participants to reveal their names, but with the media coverage of the Sweeney group’s work we realize that the project appears to outsiders as “anonymous” — even though participants, after passing our enrollment exam, know better (or should)! To meet the desires of some participants and to further clarify the non-anonymous nature of the PGP, we’re going to work on allowing participants to add their photos and/or name to their public PGP profiles. I expect it will make PGP profile pages much more “personal” and create a provocatively different scientific database!

The (sold out!) 2013 Genomes, Environments and Traits (GET) Conference is taking place this Thursday and Friday in Boston. We are celebrating the 60th anniversary of the DNA double helix with an amazing line-up of speakers and Labs.

You may watch the live webcast for free via our new channel at Fora.TV: get2013.fora.tv. Big thanks to our sponsor, Illumina, for helping to make possible the live streaming.

The hashtag for this event is: #get2013

HeLa cells with fluorescent stain.
CC0 from Wikipedia.

With such a history, we might expect extra sensitivity when researchers perform major research on this cell line. Unfortunately, this was not the case: the researchers had not contacted the Lacks family. That faux pas sparked sharp criticism from Yaniv Erlich, Jonathan Eisen, and Rebecca Skloot. A widespread outcry arose, demanding more respect for the human subjects from whom cell lines are derived. And yet we should understand the researchers didn’t do something particularly special: Joe Pickrell pointed out that HeLa’s genetic information had been published for some time through projects like ENCODE.

We are noticing the elephant in the room: cell lines were collected from individuals before the modern era of whole genome sequencing. Even if some consent was acquired, little or no warnings were made regarding potential loss of privacy.

And this is where researchers throw up their hands: “Are we supposed to stop our research? Science would come to a halt if we expect this level of consent before publishing sequence data!” The research by Gymrek et al. in Yaniv Erlich’s lab comes as an unwelcome confirmation of what George Church has been warning for years: genomes are identifiable and revealing. We may not be able to hide from this much longer: with direct-to-consumer genotyping services, research subjects will theoretically be able to detect when public sequence data matches their private genotyping data.

Specimens can no longer be treated as anonymous, they come from people. This is the new reality and we need to do better.

We can do better

The Personal Genome Project has been working to address this issue since its inception in 2005. Using an “open consent” process, we collect samples and data from volunteers who have made an informed decision to assume the open-ended risks associated with donating these items to be publicly available. We have a dozen PGP cell lines shared at Coriell, and there’s over a hundred more that will be listed soon. When the project started years ago, many wondered how many people would volunteer for the risks of privacy loss. It turns out this fear was probably exaggerated: as I write this we have over 2500 volunteers enrolled.

But we need help, especially from fellow researchers. Here’s ways you can help us create a new generation of well-consented samples and data:

• Use our samples and data – In addition to those cell lines, we have 160 genomes and hundreds of donated genotyping datasets associated with survey data for 239 traits and diseases.
  
• Cite us – At the risk of being accused of self promotion, here is the link to our 2012 paper featuring the PGP-10 and announcing other shared data and samples: “A public resource facilitating clinical use of genomes.” Also useful is the 2008 paper “From genetic privacy to open consent.”
• Enroll — If you’re willing to go public, you can also enroll and donate samples and data to the PGP. Many participants are fellow researchers, giving back to science in this manner.
• Ask people to fund us — Tell people the PGP needs support and funding. We do not currently have any government funding (not even for maintenance of our website and database). It would also be wonderful to see sequencing companies and other groups contribute genome sequencing resources.

License:CC-BY, by Brenda Gottsabend

Help wanted for cancer cell lines

In particular we could use help from a Boston-area researcher, probably with pathology experience, who is enthusiastic about joining the PGP staff as a volunteer to create well-consented cancer cell lines. We get occasional emails from participants curious about donating surgical samples but we haven’t had the bandwidth to take advantage of those opportunities. It’s a string of tasks, mostly small but doable: organizing tissue sample kits and collections, shipping instructions, tissue dissections, cell line establishment, and getting the resulting cell lines to Coriell.

If you’re interested in joining us to create a new generation of well-consented cell lines, get in touch!

Thanks again to the PGP participants who volunteered their genomes to be used by CAGI. Once the challenge is done your genomes will be linked from your public PGP profiles. We hope the PGP’s uniquely public data facilitates dialogue and leads to publicly shared methods for genome interpretation.

The website has been updated to make the csv format files containing results from google surveys publicly available. Here are links to the PGP participant survey, and the twelve trait surveys:

To help anyone interested in parsing the data, I’ve shared the Python code I’ve used on Github. There’s also a copy of the survey data as of Feb 23 there along with some demo code, and a Readme.

Finally, I did a bit of parsing of the trait survey data, combined with some features of the participant survey data (age and sex) to see if I could find anything interesting for you. The top 20 pairwise correlations below don’t look terribly surprising, but I learned some new things. For example, I didn’t know that TMJ disorder is much more common in women. (Of course, a quick web search discovers confirms this is a well known association.) This type of analysis isn’t my forte — maybe someone with more experience with machine learning can do cool stuff with this data!

(To read the table below: the first row indicates “60% of females reported a UTI, while only 12% of others reported one”.)

¹As calculated using a Fisher’s Exact test. Note that these are not corrected for multiple hypothesis testing. I think a pessimistic Bonferroni correction would demand around 1e-6 for the magic ‘p = 0.05′ cutoff.
²I didn’t look closely, but I suspect these non-zero numbers are because we have some transgender participants whose sex at birth differs from the gender they identify with (and the latter was what we have recorded on the participant survey).

This is a guest blog post from Misha Angrist, Ph.D., an author, assistant professor at Duke University, and PGP-4.

In the 7 February 2013 edition of Nature I have a commentary on genomic privacy arguing that it is time to re-frame how we think about this issue. I wrote this partly in response to the Science paper by Gymrek et al in which the authors used a combination of public genetic and genealogical data to re-identify the surnames of supposedly anonymous—or at least “de-identified”—people.

The initial reaction to my commentary from the senior author of the paper, Yaniv Erlich, was anger. He felt that I was denigrating his work, that I was implying that it had been done before and therefore it was not a big deal. Rereading my piece, I can see how he would think that; given a do-over and more space I reckon I would phrase some things differently. But I did not mean anything of the sort.

Yaniv and I had an honest and respectful exchange. His revised and measured response is here; I admire and appreciate his willingness to reconsider his initial reaction. I did my best to assure him that my goal was never to minimize what he and his team had done. On the contrary: Academic “privacy hackers” like Yaniv Erlich, Latanya Sweeney and Brad Malin are essential to understanding how secure our genomic data might—and might not—be.

What I was really trying to convey was frustration: we now have a decade’s worth of data demonstrating that genetic information is identifying. NIH (disclosure: I am a recipient of NIH funding), in its commentary that appears alongside the Erlich group’s paper, says it is concerned about this issue, but doesn’t seem all that willing to entertain policy alternatives that fundamentally challenge the status quo. The PGP, for example, is not mentioned.

Folks who study human beings are nervous. Indeed, in a phone conversation today Yaniv told me that a senior colleague of his said that his paper should not even be published lest it lead to a shutdown of public sequencing resources.

Really? So genomic identifiability is like Fight Club?

Three clarifications:

• John Lauerman is not only monitoring his medical condition, he is actively receiving treatment: see Pete Estep’s comment on my piece.
• I got Yaniv’s institutional affiliation wrong. The Whitehead Institute informed me that, “Yaniv Erlich is a scientist at, and employee of, Whitehead Institute for Biomedical Research, where his lab is located. He does not work at Massachusetts Institute of Technology.”
• I should not have implied that dead people are not covered by HIPAA Privacy. “…the medical record of a deceased patient is subject to all restraints on disclosure that are applicable to the record of a living patient.” Phew: corpses everywhere are resting easier.

Dramatic progress in re-identification has been published today in Science (Gymrek et al.), and is reported on in Wired (“Scientists Discover How to Identify People From ‘Anonymous’ Genomes”). Wired’s article features some quotes from George Church and highlights our project.

This is a guest blog post from James M. Turner, a Boston-area software engineer, freelance journalist, author, and PGP-65. James recently created a forum and wiki for discussion of topics related to the Personal Genome Project. While our staff isn’t responsible for these sites, we plan to contribute to them and hope they provide an additional place for PGP participants to find useful information and answers to their questions.

When I was 8, I read “The Andromeda Strain”, by Michael Crichton (yes, the geek force was strong in this one, even at a young age.) The book left a strong impression on me about the future of genetics, to the degree that I was writing programs in high school to convert DNA sequences to amino acids. Mind you, this was in the late 1970s, when you had to walk uphill both ways to school to save your BASIC programs on paper tape, but tell that to the kids today, I tell you…

For a while, I was sure that my future would be in the biosciences, perhaps as a geneticist. Unfortunately (or perhaps fortunately) for the biosciences, I was distracted from base pairs by the other up and coming technology of the time, computers. Although I continued to avidly follow the life sciences, I fell in love with software and have spent the last 35 years making hardware jump through hoops with clever code.

But a funny thing happened on the way to 2013. Genomics, and the obscenely steep slope of the $/genome price slide, has created another example of what trendy geeks like to call ‘Big Data.’ Big data is a challenge, because it strains the computational and storage limits of computers to analyze, but it’s also an opportunity to correlate and draw new insights from datasets that used to live in their own private silos. The PGP, microbiome atlases, health data, exercise records, phenotypic traits, diet and much more are now digital and starting to hang around in the same neighborhoods. Between the years of 8 and 50, my two passions became “two great tastes, that taste great together.”

I first became aware of the new field of personal genomics when I researched 23andMe and deCODEme for an article I wrote in 2009. As part of the research for the piece, I got my Single Nucleotide Polymorphism (SNP) data from 23andMe, while my wife tried out deCODEme. As anyone who has used the SNP services knows, there’s interesting data to be looked at, but it’s just a tiny fraction of what goes on in the entire genome. I had experienced a taste of my genome, now I wanted the real deal.

That’s what led me to the PGP, drooling into test tubes, hanging out at GET 2012, and finally receiving my whole genome sequence a few months ago. Once I had all those lovely base pairs to play with, it immediately became clear that there’s not a really good user manual for the data, a “Genome Interpretation for Dummies.” I’m a pretty tech-savvy guy, and know enough about biology to be dangerous, but I quickly found myself dealing with the subtleties of GFF vs GTF vs BED format, comparison shopping genome browsers, and coming to the realization that a “whole” genome has small holes scattered throughout it (this must be why they call it shotgun sequencing…)

One of the things I know well from software is that crowdsourcing works. That’s the entire model behind GET-Evidence, many eyes and fingers building up a larger and more useful database of gene to phenotype relationships, so that eventually a newcomer will have a wealth of information about their genome. But what’s missing right now is a place to talk about the process, learn from each other, and share what works and what doesn’t.

I started talking to the folks at the PGP a few months ago about the idea of setting up a forum and Wiki for PGP participants (and researchers, and anyone else who wants to join in) to share information, look for help, or just chat. For a number of reasons, it was decided that it would be better to have them hosted and administered outside of the formal PGP organization, and I was asked if I was interested in setting them up. I was, and have.

At http://forum.personal-genome.org/, you’ll find the Personal Genome Project Forum. It’s a place to introduce yourself, discuss the PGP, GET-Evidence, genomics, and anything else that you want to. The PGP is a community as much as a project, and people in a community should have a town square to mill around and chat.

For more formalized knowledge transfer, there’s also http://wiki.personal-genome.org/, the PGP Wiki. Hopefully, this will grow into a fount of information about the how and why of genomes. I expect there will be a fair amount of cross-pollination between the Wiki and the forums, with forum discussions turning into Wiki articles, and people discussing wiki topics on the forums. There aren’t a lot of rules at the moment (beyond the obvious about spam and privacy and civility), so the personality of the fraternal sites will evolve as people use them.

So, they’re there, they’re open for business, and with this blog posting, they’re announced. A good first start would be to drop by the forum, register as a user, and put up an introduction in the appropriate topic. There’s only two posts there right now, and it’s a little lonely. We could also really use some articles on the basics of genomic data in the Wiki, I’m going to try to contribute as I have time, and some of the PGP staff have indicated there’s stuff they’d like to write, but the more the merrier!