We have a MinION sequencer but haven't used it much. This might be a game changer if it could potentially be used to do target enrichment of exons and perhaps previously identified ddRAD loci for non-model species without having to synthesize expensive RNA baits for target enrichment. Given the translocation speed of 100s of bases per second through the nanopores, it seems like this might not work as it promises as hundreds of bases would be read before any decision can be made to reject the DNA fragment- just my thoughts.
In theory this should work on a MinION attached to a computer with a GPU:
Although we have only shown results for the GridION Mk1 this method should be applicable
to any MinION configuration with sufficient GPU to basecall a sequencing run in real time.
Early tests on systems configured with NVIDIA 2080 GPUs suggest they can keep up with a
single flowcell in real time (J Tyson, Pers. Comm.).
See also https://github.com/LooseLab/ru/issues/14
This may be one of those things that works under the right circumstances with right/clean sample? You may need to try it out yourself to see if it works for your application.
You are absolutely right, this is indeed a game-changer. However, there are some considerations to make. If you have a very small target out of a larger genome, you are effectively rejecting >99.9% of your reads, which will be rather bad for your yield. As mentioned in the preprint the yield is lower than what you would expect for a MinION run, possibly by blocking pores when rejecting a molecule. When your target size is small other approaches such as Cas9 capture would probably be more efficient. But it's possible with read until, maybe not so economically. It also depends on how many samples you sequence. RNA baits may be expensive, but so are flow cells.
The approach samples 0.4 seconds of each pore, performs basecalling, alignment and caching. For 90% of the reads a decision can be made in 1.2 seconds, often quicker, and as such for 90% of the reads up to 540 bases will be sequenced (sequencing at 450b/s). The mean length of the molecule when rejected turns out to be around 460 bases, according to Matt Loose. If your molecules are 30kb or longer (as you would aim for when preparing a library) you can save a significant amount of pore-time here. Even more so if you aim for the really long libraries in 100's of kilobases or megabases. A read of 2 Mb (do the math) is keeping a pore occupied for a very long time. Note that also as soon as a read gets rejected the motor is moved and it won't be caught by the pore again.
In theory this should work on a MinION attached to a computer with a GPU:
This may be one of those things that works under the right circumstances with right/clean sample? You may need to try it out yourself to see if it works for your application.