This blog post was written with Roy Ronen from UCSD (@roy_ronen). We thank Dina Zielinski (@dinazielinski) for super helpful comments!
Oxford Nanopore was very accommodating, allowing us to spend some time in their Cambridge, MA office trying out their MinION sequencer after ASHG. We hung out with them for almost two hours and asked many questions, some of which they were able to answer. Most importantly, we got to run a pre-processed phage λ library ourselves and get a feel for some parts of their sequencer.
Our MinION actual run. The computer shows the
Gulper Grouper program that controls the device and presents the signals from the nanopores.
Guessing the capacity of a MinION
Although we did not see any FASTQ or base-calling files, we can do a back of the envelope guesstimate of the MinION sequencing capacity based on the various technical details [All of these details are already and previously appeared online (e.g this excellent blog post by @pathogenomenick)]
MinION has approximately 6 hours of total operation time; the DNA fragments pass through the Nanopore at an approximate rate of 10bp/sec; there are 512 ASIC sensors each connected to a single nanopore and the flow-cell costs $1000. Therefore:
Total sequencing capacity per flowcell = 6 x 3600 x 10 x 512 = 110.6Mb
Cost per megabase = 1000 / 110.6 ≃ 10$
These numbers assume full capacity of the sequencer, meaning that all the nanopores are fully operational all the time. Therefore we believe them to be somewhat of an overestimate of the true capacity.
Let’s place this guess in the context of the famous NHGRI sequencing costs slide:
Black: NHGRI sequencing costs. Cyan: Our estimation for ONT.
If the error rates are reasonable and the promises of long sequence reads can be delivered, MinION is going to put up a strong fight against PacBio (see a Twitter discussion from Oct 13 PacBio prices here). Indeed, the price of PacBio per Mbase is about five times lower. But PacBio also costs an additional >$500,000 for the sequencer, maintenance, and valuable real estate. You need to do a lot of PacBio sequencing to amortize these expenses.
However, one important take-home message is that MinION (and presumably its GridION scale-up) is far from being a threat to Illumina, and this is before talking about error rates. ONT is almost 2.5 orders of magnitude more expensive than Illumina. In addition, Illumina sequencing benefits from a very strong data analysis ecosystem, with mega-projects such as 1000 Genomes, that developed full bioinformatics suites tailored to the specification of this technology. To pick a fight with Illumina’s home field advantage of medical sequencing is going to be extremely hard.
MinION is not a sequencing platform. It is a sequencing sensor.
The analysis above overlooks one crucial component of the MinION technology: it is so f***ing small! It felt approximately the same weight as our cellphones and costs around the same order of magnitude.
What does this mean?
Our feeling is that the MinION is a sequencing sensor, not a sequencing platform. Bye-bye 110V/220V electricity, -20C freezers, AC systems, PCR machines, data-centers, real-estate, and liquid nitrogen. Hello electronic boards, microfluidics, and standard I/O. IMHO, the killer app of ONT is EMBEDDING their technologies in bigger systems: smart water pipes that sense the microbiome, automatic quality checks for the food industry, sewage epidemiology, and even biometric authentication systems.
In the sequencing sensor domain, raw sequencing error rates or throughput are less of a concern. The technological focus becomes seamless integration and the ability to count molecules.
There are of course many challenges to overcome before reaching the full sequencing sensor stage. But this uncharted domain will bring exciting developments on both the research and business perspective.
[Nov 11, 2013 Update]
ONT has written to clarify these points: “[W]e were running machines in this range [10 bases/sec] at the demo event however this is now on the lower side of the potential range which has the potential to reach tens and eventually hundreds of bases per second. We have not given guidance on this at the moment because we will be developing this further as time goes on and evaluating variety of speeds in the field with MAP. But the range that you quote for one MinION use is not it’s upper commercial limit by any means. [The] $1,000 per flow cell is the MAP pricing, we have not confirmed commercial pricing yet although believe it is likely to be in this range.
[They also asked to add the following text from their website regarding GridION]:
For those considering the minimum resources that may be needed to acquire and operate a GridION system, they may assume that at the time of commercial launch this will be lower than other commercially available systems on an absolute level and that the cost per base will also be lower than other systems. The system is modular so customers may add nodes to their installation. No accompanying server will be needed as each node contains the required local computing hardware.”
Appendix: sample prep
The first version of the sample preparation kit has five tubes:
1. Control DNA: lambda phage to test the sequencer.
2. Tag mix: Transposase to fragment and attach the loop and adapter pair.
3. Repair mix: to bind the loop to the DNA fragments, which allows both stands be sequenced
4. E4 enzyme: to slow down the passage of the DNA fragment through the pore.
5. Ep buffer: sequencing buffer.
It seems that Oxford Nanopore put a lot of effort into making MinION a truly portable sequencer. According to ONT, the sample preparation has only four steps with no PCR amplification and the library prep is almost entirely conducted at room temperature (RT) in ~1 hour (there is a single 10min heat inactivation at 70C, but ONT is actively thinking about how to eliminate this step). Starting the run was quite rapid: we just plugged the sequencer to a USB3 cable and ran two (Python!) scripts through the GUI: quality check of currents passing through the pores (~40s) and quality check of chemistry (3 min). The second step relied on a DNA molecule that is shipped inside the flow cell and used to QC. Then, we pipetted 300ul to the pore on the USB device and started to accumulate data (again we don’t know anything about its quality).