[waiquoo]

Researcher working on a different approach to nanopore sequencing here. The Minion is really interesting technology, but early reports basically indicate that it's essentially useless in it's current form. One of the issues is that the basecalling algorithm relies on a noisy, two bit signal. Apparently it works okay on trained sequences, like lambda DNA (that's where the 60-85% accuracy comes from). But when used to sequence untrained DNA, the accuracy drops off significantly (<10% accuracy, http://onlinelibrary.wiley.com/doi/10.1111/1755-0998.12324/a...).

There is lot's of room for improvement though. All of the commercial nanopore tech is based on biological nanopores, which have the advantage of having very straight-forward to fabricate. But they are limited to the ionic current signal, which is very noisy and weak. Once these companies start introducing solid-state devices though, things will begin to get very interesting as alternative signal transduction mechanisms come into play.

[troymc]

It sounds like they are reading several DNA strands in parallel at the same time, and each output-sequence has noise. It seems to me the problem then becomes one of finding the most probable "signal sequence" given all those noisy output-sequences. Oh, and it also sounds like you wouldn't know which letter is number 1, which is number 2, etc. Is that right?

It seems like a fun problem in information theory. Can you point us to some articles or papers about current approaches to solving it?

[waiquoo]

Winston Timp has done some interesting work in this area (http://www.sciencedirect.com/science/article/pii/S0006349512...). Basically, by training a hidden Markov Model, you can get the most likely sequence from a noisy source.