By Monica Heger
Vying for an edge in the benchtop sequencing market, Life Technologies and Illumina have recently stepped up their campaigns to convince users of the superiority of their respective platforms, the Ion Torrent PGM and MiSeq.
This summer, the sequencing of the Escherichia coli outbreak strain in Europe offered a first look at sequencing data from the two platforms (IS 7/5/2011), kicking off a series of application notes and presentations from the companies.
Following the initial comparisons of the platforms, Ion Torrent in early August released an application note touting improvements in the PGM that would yield average read lengths of 163 base pairs, showcasing a 265-base perfect read. At the same time it claimed superior accuracy compared to the MiSeq, as well as better scalability.
In response, Illumina in late August published an analysis of its own, addressing each of Ion Torrent's claims. The company asserted that Ion Torrent misrepresented the data by omitting indels from its accuracy comparison, and claimed that the MiSeq was in fact the more accurate of the two platforms.
However, which machine is more accurate is dependent on a number of factors, including the tools used to analyze the data, how the data is interpreted, and a researcher's specific application, noted Nick Loman, a bioinformatician in Mark Pallen's research group at the University of Birmingham in the UK, who tracks next-gen sequencing platforms on his blog.
"The devil is completely in the details," he said.
Each platform is prone to different error profiles, said Loman. "The Ion Torrent is more likely to generate indel errors in homopolymer regions, whereas base substitution errors are the predominant error type in Illumina data."
Depending on the application, one error type may be less of an issue than the other, but the best scenario would be a "random source of error, rather than systematic," he said. Systematic errors cannot be averaged out by increasing coverage, whereas random errors can. Other Illumina sequencing platforms have produced more errors in regions of high GC content, but it is not yet clear whether the MiSeq has that same error profile or is more random, he said.
Both companies also biased their presentation of the data in order to show their respective platform in the best possible light, he said. While that is not surprising, "it makes sense for there to be an independent evaluation of the datasets to try and get a fair comparison."
For instance, Ion Torrent claims superior accuracy at base 150, with a 2.99 percent error rate compared to the MiSeq's 11.2 percent error rate. Those differences, according to the company, put it in a better position to generate even longer read lengths in the future, because accuracy does not trail off. Loman said that while that may be true, at the moment, the statistic is not very meaningful because so few reads are generated at that length.
Additionally, he said, Illumina presents an over-inflated view of the PGM's indel error rate. Evaluating sequencing data from E. coli, Illumina said that the PGM generates thousands of indel errors. However, said Loman, "when you do consensus calling, you don't see that many indels." In fact, Ion Torrent claims 32.
[ pagebreak ]
One way to get past the companies' hype is for researchers to analyze the data themselves. "To be fair to the companies, they have released the datasets," Loman said, "so there's nothing to stop anyone from doing the analysis themselves."
Several researchers have begun doing just that, including Monkol Lek, a bioinformaticist and PhD candidate at the University of Sydney who has posted an analysis of the Ion Torrent data on his BioLektures blog, and EdgeBio, which has published a comparison of datasets from the two sequencers on its blog.
Keith Robison, a senior scientist at Infinity Pharmaceuticals who covers sequencing on his blog Omics! Omics!, told In Sequence that there is the "danger of an apples-to-oranges comparison" in trying to compare accuracy between the two platforms.
What's important, he said, is knowing the types of errors each machine produces and whether those types of errors and error rates are acceptable for a specific application.
Robison described the flurry of application notes that the companies have been releasing as part of a "fierce battle" as the companies vie for a slice of the benchtop sequencing market. "The companies' hope is that every pathology lab in the country will have one," he said.
Amanda Murphy, an analyst with William Blair and Company, agreed that the market for PGM and MiSeq could be large, "encompassing at least the 6,000 CE machines that have not yet been replaced by next-generation sequencing but are used for higher-throughput applications," she wrote in a recent analyst report. Longer term, the market could extend "beyond the life sciences community into agro-biotechnology and clinical diagnostics."
According to Murphy, aside from accuracy specifications of the platforms, scalability will play a major role in determining which firm gains an edge in the market. Whether Ion Torrent can continue to scale the PGM as promised — increasing its throughput by 10 times every six months — "will be a key driver of how much market share each machine is able to secure," she wrote.
Ion Torrent has been claiming superior scalability versus the MiSeq, a claim that remains to be proven, said Loman. "There is good evidence that it's scaled well so far," he said. For instance, Ion Torrent has increased its throughput with each new chip release, and users have generated even more data than its specifications indicate, he said.
However, given Illumina's history in the sequencing space, Loman said it would be safe to "assume they've got plenty of tricks up their sleeve."
"The question is, when the MiSeq comes out, where will Ion Torrent be, and where will MiSeq be?" he said.
Have topics you'd like to see covered by In Sequence? Contact the editor at mheger [at] genomeweb [.] com.