Thursday, February 24, 2011

Reflecting & Responding on Today's Comments

It's nice to see my piece on the Ion Torrent 318 chip get a lot of comments, as well as being tweeted and some additional comments at SEQAnswers. I love getting comments, especially those that constructively challenge my analysis.

Two commenters threw some cold water on my enthusiasm, pointing out that the 318 is not scheduled to be released until September, and then only to early access customers. I'll confess to not asking careful enough questions on the timing & letting my desires get ahead of me. Ion Torrent will need to deliver these chips on time and working at specifications in order for scientists to believe that there really will be the promised regular upgrades. MiSeq is vaporware at the moment also, but Illumina has a strong track record for rolling out instruments.

One commenter took me to task for suggesting that Ion Torrent will simply roll right over Illumina. I don't mean to suggest this will happen, only that Ion Torrent may offer a very stiff challenge. If the PGM chip upgrades, sample prep improvements and read length improvements come as promised, then the PGM may well march up through Illumina's product line. That's not to say Illumina will remain static; they have clearly demonstrated great skill at pushing their platform to amazing abilities. The MiSeq may offer a new path to greater performance; there has already been speculation over at SEQAnswers on possible paths forward. For example, will the faster cycle times mean an opportunity for longer read lengths (if the reagents are not stable over very long runs)? Will Illumina roll out a multi-flowcell version of the MiSeq, enabling very high-throughput rapid-turnaround sequencing? I'd put more money on the latter than the former. Another thread has surfaced rumors that a HiScan upgrade is imminent which will give it HiSeq-like capabilities; for groups running both arrays and sequencing this could be useful. How much denser can Illumina pack the flowcells? Only time will tell. But, as I pointed out before, longer reads will have limited impact on overall throughput if the two reads are overlapping (though accuracy may be improved).

On the other hand, I do see as excessive cynicism (bordering on editing the truth) that AGBT had only Ion Torrent talking about Ion Torrent; given that two talks (Bolen from NCI and Nusbaum from the Broad; the Bolen talk will apparently be at ABRF as well) described actual experiences. Granted, neither was really independent of the company. It is critical for Ion Torrent's reputation that data start appearing in public spaces, especially data from regular customers.

Ditto for the experimental protocols; one commenter had a question about amplicon sequencing which will be important for Ion Torrent's success. The sooner applications are adapted to the system, the sooner customers will be buying machines and chips. Of course, a great strategy IMHO would be to put some sequencers in the hands of genomics bloggers.

As far as the question of reads per chip increasing faster than the number of sensors per chip, my understanding is that Ion Torrent expects that better sensor loading and utilization. Again, how vaporous is this? That's hard to tell -- and why feedback from actual users is desperately needed.

On the question of amplicon and read length for Ion Torrent, those would seem to be different fish (though I don't have direct experience). Amplicon length will be a matter of the emPCR conditions. I'm unaware of any signal amplification in the 454 chemistry -- I thought the signal cascade was strictly linear. But, again that is outside my expertise and I would enjoy being educated on the subject.

With regard to whether emPCR can be converted from a liability to a non-issue, it's reasonable to be skeptical. But, it isn't obvious why emPCR can't be compressed in time (though I haven't done emPCR, so that's a stretch for me to comment on). In my discussion with them, they pointed to cutting the number of cycles as one time savings. In any case, how hard as anyone tried to really optimize emPCR for speed? Only on the 454 might it have been a serious concern, and the cost of a 454 run is not going to encourage many folks to push for many runs a day. On the other hand, I don't see it as "disingenuous" the comparison I made; I've been at a company that had techs running PCR (again, conventional not emPCR) day in and day out, and any company that runs production Sanger is in a similar boat. Indeed, any big SOLiD or 454 shop must do this as well. What will the nature of the new emPCR kit be? I have some speculations, but that's a whole 'nother post.

One other surprise today from Ion Torrent is the announcement that they have sequenced the genome of Intel co-founder Gordon Moore. One claim from this is that the coverage was more uniform than with other platforms. Obviously, either publication or release of the data is needed to vet this claim. Even if they used 318 chips, this was an expensive demonstration project (I don't know the actual coverage; I don't have a subscription to In Sequence) -- even at 10X coverage it would be $20K (retail) in chips. Apparently this was an exclusive for GenomeWeb's In Sequence, as Ion doesn't even seem to have a press release out. I hope they plan to publish this in a journal soon, or if not make the primary data available for further study.


Doron said...

It was not an exclusive for In Sequence, Rothberg announced the sequencing of Moore's genome during his talk at Molecular Medicine Tri-Conference

Anonymous said...

Its an electrical 454 machine. The density of beads is higher. Moores law does *not* apply to this system as the density of transistors isn't limiting - what limits is the density of beads and kinetics of the chemistry around those beads. It can never be long read because pyro-like chemistry cannot deliver that on a multi-molecule format.

The main positive change is the shift towards smaller cheaper boxes and the attempt to make it accessible to benchtop users.

The rest is hype. As one commenter said, AB have pulled their classic trick of making a virtue out of a necessity in the marketing (as they did with 2 colors on SOLiD), aided by the chutzpah of Dr R. Only the weak minded fall for old Jedi mind tricks like that.

Anonymous said...

Why is Life falling into a marketing trap again. Customers want a fast turn around solution for long read in a possible molecular diagnostic lab or clinical interface research unit. Why is Life now showing how they do complete genomes on a larger chip ? RNA Seq or even exome ? That's just not the aim here.

Start working on longer reads please, yesterday they even talked about 400 bp in the R&D lab (how about the reads TODAY !).

How can anyone measure the delivery of a chip, R&D'ed, supported, produced withing a relative 7 months from now ? Anyone who has worked internally at Life for the last says 10 products, know that they have never been able to match delivery time. In some successful instances, best specs had been botched in order to meet deadlines... (and bonus'...), the beauty of GE like management.

So I hope they get their act together, as always, way too many sheriffs in the coral.

It's fine to aspire to Illumina killing with PGM, I suspect they will also kill SOLID if that happens. Again, people who look into PGM want nothing to do with massive data overload, they want simple, give them simple.

Another area Life might want to focus on, a lot of massive centers out there will want a second or third reader to do some validation (it's been proven most system have some bias). This is a good little niche but Life might want to make the case that IT does give a different quality of reads. Long gone are the days that anyone would simply take their words for it, show us some data.

Right now, I have seen reps tell me that the datasets shown were amazing, based on a single slide conclusion from two AGBT talks. There is a problem there... I actually saw those talks in person, and I would consider these slides to side on marketing BS for now...

Anonymous said...

In some ways it is too bad that Ion Torrent was bought by the makers of SoLID rather than 454, as the Newbler assembler is what is needed for the Ion Torrent reads.

My 3rd-hand impression is that the Ion Torrent reads are currently too short (70-100 base pairs) and have too high an error rate (2-5%) to push much against Illumina or 454. Ion Torrent was pushed to get their product on the market before it was fully debugged. They might have been better off keeping it in development for one more year.

Anonymous said...

Life Grand Challenges

Discover more about the Life Grand Challenges and how you could be the winner of a $1 million prize.
Challenge background: scalability—twice the bases

The idea behind Ion Torrent Technology™ always has been to bring the power of semiconductor scalability to the life sciences. The Ion 314™ Chip is the first in a series of products that leverage decades of semiconductor technology advances, enabling us to bring the entire semiconductor design, fabrication, and supply-chain infrastructure—a trillion dollar investment—to bear on the challenge of sequencing.

If we have 1.4 million sensors on the Ion 314™ Chip, we should be able to get more than 1 million reads. The winner of this challenge will have to provide a 2x improvement in the number of high-quality bases across our available chips. The benchmark accuracy that challengers must beat will be the best yield achieved by the Ion Torrent™ labs, as measured in a validation event.
Challenge background: speed—twice as fast

Speed is critical to scientific discovery, particularly as sequencing technology moves from the research lab to the clinic, where decisions need to be made in hours, not days.

This challenge requires participants to prepare a sample in half the current record sample prep time set by technicians in Ion Torrent labs. Currently, the Ion Torrent record is about 10 hours, with each step completed back-to-back without pause, using all the best processes. Selected entrants will compete in a validation event against an Ion Torrent team running the most current protocol.
Challenge background: accuracy—twice as right

The accuracy challenge is unique in that it is based purely on software improvements—specifically, twice our internal accuracy record at the time of submission.

For example, if we are producing 40 million bases at a given accuracy, the challenge would be to produce 40 million bases at twice that accuracy, at the same or better analysis time and using the same server configuration as the Ion Torrent product uses.

It is not necessary for the programmer to know anything about biology or to have access to an Ion PGM™ sequencer to win. Any programmer, anywhere in the world, can compete and win.

If this is not a marketing scheme, I don't know what it is ? Keith, I am curious, do you feel 'used' by marketing because I find most of these tactics questionable. Again, folks are waiting for basic basic data, yet Life will give millions for unreachable goals...

Anonymous said...

454 signal is "amplified" because they turn the release of PPi into a light signal, and that has higher SNR than trying to measure H+ directly. Has anyone realized that any buffering in the solution directly decreases the ability to measure H+? Ion's sequencing is teetering at the edge of a cliff, you should be surprised it works at all. Have fun supplying ultra clean water to push out those reads. Marketing can make anything complicated sound simple.

Anonymous said...

if that happened, nothing in biology is right