August 10, 2009 | Working on a single instrument in a single lab, a single research associate generated the first single-molecule human genome sequence in a single month. According to Stanford professor Stephen Quake, a.k.a. “patient zero” and co-founder of Helicos Biosciences, his group’s success is proof of the growing democratization of genomics. Kevin Davies caught up with Quake on the eve of his landmark personal genome publication. (See the accompanying Bio-IT Worldstory on Quake’s personal genome.)
Bio-IT World: This must be very pleasing from six years ago when you managed to sequence all of five bases?
QUAKE: It is. I feel we’ve managed to come full circle.
You talk a lot in the paper about the democratization of gene sequencing.
There’s a table in the supplement which indicates the effort that’s been needed to sequence human genomes up until now. Our work is important at this time in that this is the first case you haven’t needed a genome center to sequence a human genome. What we’ve shown is that you can do it with a pretty modest set of resources—a single professor’s lab, one person doing the sequencing, one instrument, lower cost. Those are all order-of-magnitude improvements over what’s been published recently.
That being said, as you’re no doubt aware, the DNA sequencing industry is certainly competitive. Everything is moving fast, very much in flux. All the manufacturers are improving their platform by a factor of two per year. I’m just saying, at this point in time, Helicos is the best platform, and they’re going to be in a dogfight to try to keep that title -- which is good for the scientific community.
There’s very little mention of Helicos, which you co-founded, in the paper. Did you deliberately set out to keep this a separate effort?
Yeah, it’s complicated. One of the reasons is the conflict-of-interest rules of my institutions. I’m part of two institutions – both Stanford and the Howard Hughes Medical Institute. They have almost orthogonal, non-overlapping conflict-of-interest rules, very constraining. One upshot of that is I’m not allowed to collaborate with a company… It’s much more strict on the Hughes side, it’s like one of the Ten Commandments.
You didn’t buy the HeliScope outright, you collaborated with several other faculty?
Exactly. The machine was purchased by the stem cell institute at Stanford. The purchasing process was very transparent. There was a call for bids, it was done in a very open manner... So I benefited from collaborating with those guys. The reason they bought it was not to sequence my genome, but to sequence cancer, tumor stem cell genomes. That’s what’s up next. Mine was just to practice, to show that we could do it and to get the informatics into place.
The supplementary information put the price of your genome at $48,000. Can you elaborate?
Those were just the reagent costs. The amortized machine cost is about another $10-20,000.
Why didn’t you name yourself in the paper as Patient Zero?
Well, you know, we wanted to retain some semblance of dignity for the scientific literature! It’s really irrelevant for the purposes of the paper.
Your grad student wrote the variant calling algorithm. Was that out of necessity?
In fact, Helicos wrote a mapper but not a base caller. We used their mapper, which is tuned to the error profile of the instrument. All the mapping softwares have been written with particular instrument performance in mind. For example, Maq and ELAND are written basically for the Illumina platform, where the dominant error is substitution. For Helicos, the dominant error is deletion, and that has consequences for how you do the algorithm. We used the Helicos mapper [IndexDP], but then, all the base callers are tied to the mapping software. So ELAND and Maq will call the bases, but it’s all linked into how they do the mapping. So we ended up writing our own base caller.
If you look through the literature, the way base calling is done in the other published genomes is still rather ad hoc. I wouldn’t say there’s real consensus. People put on arbitrary conditions. Some people like to use a priori knowledge of human variation. A database like dbSNP can be used, but we wanted to take a very rigorous approach where we ignored everything that is known about human variation, and just try to call the genome based on the data. Then use what’s known about human variation to validate the calling. So that was another reason.
The genome coverage was 90%. Would you get higher coverage with more reads?
There are very repetitive parts of the genome that don’t map well. Most people aren’t mapping to the whole thing. The Chinese one was also 91-92%, something like that.
The paper talks about the deletion errors...
Yeah, that’s the primary source of error – deletions due to these ‘dark bases.’ One of the reasons this is an interesting result for the academic literature is: Is it possible to sequence the human genome with reads that are a little shorter and different dominant error mode than you have on other platforms? We show that it’s definitely possible.
You didn’t get into this in the paper, but you’ve done some preliminary analysis of genetic conditions. Did you use the Church lab’s Trait-o-matic program?
That’s right. George [Church] was very kind, and ran it through Trait-o-matic. That’s where we got a preliminary annotation. We didn’t discuss that in the manuscript, which was more about the technical aspects of the sequencing. We’re preparing another paper on the annotation. In fact, my medical colleagues have gotten really interested in this. There’s a small army doing a hand annotation for things that aren’t covered in Trait-o-matic yet, like pharmacogenomics. That’s going to be quite a lot of fun going forward.
You’ve said earlier this year (in the New York Times) that your daughters have pretty severe peanut allergies. Where do you start looking for that?
Yeah, that’s a really interesting question. The genome is maybe not the answer to that, right? The immune system has this interesting property that it rearranges the genome. All the immunoglobulin genes are rearranged in B cells and T cells. That’s more of an epigenetic question, one I’ve got a great interest in. We published a paper this spring in Science describing how to sequence all the expressed antibodies in a model organism, zebrafish in that case. I’m taking a more direct epigenetic approach to these questions when the immune system goes haywire. It’s possible classical genetics will be helpful. My personal opinion at the moment is the way these technologies can be used is to measure immune repertoires and understand what’s happening from a more physiological point of view.
What other research uses do you foresee using the HeliScope?
We already have three more genomes in the can related to leukemia and cancer. We’re neck deep trying to analyze those and understand what they mean.
After a tough 2008 for Helicos, this must be a very timely publication?
It’s hard to say whether there will be any impact. It’s kind of a David vs Goliath battle. There are four commercial platforms out there right now, and three of them are billion-dollar companies. The fourth is Helicos, which is a scrappy little bunch -- they’re trying to hang on! I think they’re fantastic, and I’m hoping they’re going to end up at the top of the heap.