Tuesday, June 21, 2011

A first peek at data from an Ion 316 chip

      Ion Torrent has made available an E.coli DH10B fragment dataset for the 316 chip, which is expected to be generally available early next month.  I've theoretically had access to the data since Saturday, but a series of events (computer-free weekend, a crashed home computer, a personal day & a good but unbloggable conference) have meant that ambitious plans to analyze it are still in progress.  So, what I discuss below is either from Ion's provided information or other sources, so take it with appropriate caution.

       The first statistic of note is that their dataset has 1.69M reads, with 1.53M of those >=50 bp long and 1.07M 100bp or longer.  No enormous gains in read length in this dataset, though the curve might be shifted towards longer (one of the many analyses still being done).  Now, the question immediately arises as to how typical this run is; their public dataset from the prior 314 chip with >400K reads is clearly an outlier 
      However, the Broad Institute has been saying publically that they are getting over 2M reads from their 316 chips (the Broad is an early access customer).  Also, it is interesting to note the loading plot for this run, as it has two regions of essentially zero sensor loading which appear to be perfectly circular. My interpretation is that these are two bubbles that occurred during bead loading, meaning that Ion is actually showing (perhaps inadvertantly) a less than stellar run.  On the other hand, perhaps it was a really good one that just has these flaws.  

    Jonathon Rothberg declared at a recent conference that the PGM would hit readlengths of 400 by the end of year, but as mentioned before any progress in this direction is subtle.  However, Ion has stated they plan to work on performance by focusing on one area at a time, so perhaps an increase in length will come next.  Of course, to go much longer than the current will require increasing fragment size, an issue I've bemoaned.  On the other hand, if the same efficiency applied I could expect to get 250K reads with the 316 chip; given that the chip is expected to be only a few hundred dollars more that might be a reasonable price to pay.
   Release of the 316 on time and at specification is an important step for Ion to build confidence that they really can deliver regular performance updates.  But, even this demonstrates only so much.  While Ion has greatly ballyhooed the semiconductor nature of their platform and Moore's Law, I've been finding (via correspondence) that skepticism on this point is not uncommon.  Moore's Law, after all, has been largely driven by decreasing feature sizes on chips, enabling more transistors to be packed in.  Ion likes to boast they are using relatively ancient chip technology, so transistor density isn't an issue.  Indeed, the 318 chip is reported to use only 2 transistors per sensing element, versus 3 for the prior chips.  Ion's issues will be around how densely they can pack wells before significant cross-talk occurs, or simply how much tighter they can pack their sensors before the bead sizes simply must be reduced.   Of course, longer reads and even more efficient loading and bead prep are other avenues, but with more limited headroom.  Changing the bead size will be a major ravine that Ion will need to cross, but it isn't clear when they'll be forced to make that change.


NGSman said...

Having used the 314 extensively, we are skeptical that the 316 can reach these marks, even witht the Broad's claims. Read length is the major shortcoming with these chips....

Jmaes@cancer said...

I really want to see Paired end data coming off the Ion Torrent, again something thay have said will be coming.

Will Illumina or Ion win the race for clinical hearts and minds is the big questions for me. Whichever system first gets into clinical labs and is approved for diagnostics could be the straw that breaks the other camels back.