Since the start of Oxford Nanopore’s MinION early access program (MAP) in June this year, the research community (at least those of us not lucky enough to be part of the program) have been waiting to see the first results. Would this handheld sequencer marvel actually work? What kind of data would it produce? What kind of read lengths could it achieve? How about the error rate? We’re just now starting to get some of these answers. The verdict is in – while flying cars are still confined to science fiction, we now have a handheld sequencer! (Provided you can also manage to hold on to your laptop at the same time – a computer is required for the MinION to run.) While precise numbers are hard to come by, participants in the program seem to be routinely generating around 200Mb to 300Mb per run, with some getting as much as 500Mb. Internal runs at Oxford Nanopore are reported to be as high as 1Gb.
The other major positive outcome is that the reads seem to be quite long – as least as long as PacBio and in some cases substantially longer. There have even been a number of reports of 100kb+ reads. The only real negative is the one people were expecting – high error rates. The numbers seem to fluctuate quite a bit, with reports coming in between 10-40%. Even more important than the raw error rate, however, will be the error model of the system. PacBio suffers from a high error rate (~15%), but the error model is almost completely random. This means that with oversampling (multiple reads of the same molecule or of different molecules from the same genomic region) can yield very high quality consensus reads. So far, the MinION error model isn’t quite clear. Supporters talk of 100% accuracy when aligning reads to a reference, but we haven’t yet seen any clear claims of a stochastic model or of low error with de novo assembly. There are some slightly worrying hints – many of the very long reads don’t seem to align to any reference. MAP participants talk about improving alignment scores by bioinformatically chopping up their long reads into smaller chunks prior to alignment, although in this case, the ‘small’ chunks are still a quite sizable 5kb or so. This method gets around the fact that the nanopores tend to produce errors in long stretches. The base caller relies on the results of the preceding bases, so one incorrect call can have a cascading effect. This will likely be mitigated in the future as the base-calling software improves and as new alignment algorithms are written to take advantage of the ultra-long reads. In addition, the MinION has already seen chemistry improvements, leading to lower error rates.
Recently Oxford Nanopore has started talking about their next sequencing platform, the PromethION. This system will in essence be a higher throughput version of the MinION. The MinION operates with a single flow cell (with each flow cell having ~500 nanopores). The PromethION will operate with multiple flow cells, and up to 100k channels, allowing for substantially higher output. Oxford Nanopore’s CTO, Clive Brown, recently stated at Nick Loman’s Balti and Bioinformatics “On Air” Google Hangout (archived video) that it would be able to generate ~300Gb of sequence when first released, climbing to over 1Tb per run over time. With this kind of output, nanopores would start having a much bigger impact on the overall market, likely putting pressure on Illumina’s flagship HiSeq sequencing platforms. Oxford Nanopore will be releasing more information about the PromethION at ASHG next week, so stay tuned!
And for those who are curious and have long memories, the GridION, which Oxford Nanopore doesn’t seem to talk about much anymore, is a system for connecting and coordinating the output of multiple sequencers. In theory it could operate multiple MinIONs or multiple PromethIONs (although the form factor they show on the web doesn’t seem to match either of these two systems).