Wednesday, May 23, 2018

Miscellaneous & Disorderly Thoughts on the Eve of London Calling

It's the night before London Calling. I hope to post Thursday, but an after-meeting report won’t be until nest week - I must dash on Friday fir a slightly insane/exhilarating routing to meet my family in Florida for the holiday weekend. Exhilarating as I will have a layover in one of the ancient capitals of Europe, Lisbon, which I’ve never visited. Insane, because it’s a 12 hour overnight layover. Anyway, between the challenge of covering Oxford Nanopore's expanding reach of products and applications and being sleep-addled from taking the redeye flight I'm going to throw out a bunch of thoughts without really trying to fuse them into a coherent narrative.

I’ll buy a drink or dessert for whomever can silk Tiny a Tim into the soundtrack before Hans Jehlen’s  talk. Another for whomever snaps the expression on Gordon Sanghera’s face when this happens.

As noted before, PromethIONs are shipping vigorously. It does appear that multiple users have experienced low yields on their first few flowcell, but quickly improve to 50Gb-plus yields. But what is causing there to be a learning curve?

Will VolTRAX ever be more than a shiny toy? I’m happy to be proven wrong, but it seems to not be routinely used much - given the simplicity of the rapid 1D workflow, VolTRAX isn’t any easier. ONT has promised additional kits for ages; what will be promised this tome? Perhaps if it doubled as a thermocycler it could be more popular for home labs - though still pricey versus some available instruments. Slightly off-topic:  what’s the current betting line on whether Bento Lab will ever deliver a product? Their centrifuge-gel box-thermocycler kit would be great. Perhaps they, well, boxed themselves in with too low a price point?

On that general subject of ONT promises, I often regret not tracking delivery times more carefully. ONT delivers most of their pronises, but often with substantial delays. A recent example: the shear-free ligation kit was originally slotted for early April but still isn’t in the store.

The new MinKNOW GUI is a spectacular improvement. I had at times contemplated writing a piece slamming the old one. In my darkest moods, the working title was “How do I hate MinKNOW?  Let’s count the ways”. That’s truly a dead idea now! Not only does it look professional, but it’s now much more fire-and-forget for an experiment. There’s still too many cryptic product codes and Inwoukd have made a couple of tweaks to colors in the status displays, but the confusing business is mostly gone. Pore states have been compressed to a small number - and they have labels that are meaningful outside of ONT engineering meetings!

The new MinKNOW also has two upgrades -- better voltage control and a more aggressive unblock scheme -- aimed to boost throughput. Many people reporting a noted improvement, Pore Camp Vancouver delivered a new overall record of 17 gigabases for a MinION flowcell - and even at starbase we got a spectacular run for us (stopped after 24 hours just shy of 7 gigabases).

SmidgIONFlongle. I’d love to be wrong, but I’m betting that hopes of the goodie bag will include a Flongle or SmidgION will not come to pass. I expect the demo area will have at least one running Flongle demonstration, but given that ONT hasn't made any further noises about this I suspect the shipping date of late this month is going to slip.

GridION is a bit of the dull workhorse of the current lineup. It doesn't have the huge yield of PromethION nor the form factor wow of MinION. Other than occasional reports of instruments showing up dead on delivery (and being immediately replaced), they don't make much news. Clive Brown has hinted on Twitter that GridION sharing flowcells with MinION might not last forever -- perhaps a concrete announcement will be made? And would that be a switchover to PromethION flowcells or yet a third flowcell type? I do still love the idea of a GridFlongle -- GridION body set up to run as many Flongle flowcells as the form factor comfortably allows. With the increasing yields, Flongle is likely to be a better fit for many applications than MinION.

MinIT is the informatics gadget to support basecalling and downstream analyses. Another toy I expect to be demoed on the floor. When will they actually ship?

Zumbador is the concept for a device to extract DNA from biological specimens and generate a library for loading on a MinION. It's a pretty tall order, even if you exclude sample types (bone, plant, gram positive bacteria) for which simple lysis isn't sufficient. Will there be actual data from Zumbador this year?

The field kit has lyophilized reagents for preparing rapid 1D libraries, and ONT has (if I remember correctly) been beta testing these with select users.  Full launch would be very exciting for field applications and for clinical applications in developing nations.  But also expect such kits to be attractive to the hobbyist and educational markets.

ONT has described various uses of Cas9-based capture for sample enrichment or even library-free target counting applications ala Nanostring.  This also was undergoing a limited beta release -- updates will be eagerly watched.

Totally off topic, but when I'm in England this bubbles back into my head.  The English colonists in the Americas were pretty shameless about borrowing town names from back home -- New England is particularly heavy on these (occasionally prefixed -- if ONT wanted to have a New England conference perhaps we'd have a New London Calling) but it's also common throughout the other original colonies.  I grew up in Chester County Pennsylvania, which abuts Lancaster County and sports such town names as Malvern, Oxford and Devon (as well as yet another New London).  But this leads to the question -- what explains the town names that don't reappear on our side of the pond?  I don't know of any U.S. towns named Woking or Rotherhithe or Barking (though I haven't checked). Geography of where settlers came from? Places that really didn't matter when the main colonization was taking place?
 
What's the youngest level of education anyone has pushed MinION into?  Any pre-K sequencing-and-tell going on?

Beyond the current platform, what will Clive talk about?  Detection of proteins or other analytes that aren't nucleic acids?  DNA synthesis?  Real data from solid state pores?

Anyone using direct RNA for bacterial transcriptomes? Nature wasn’t kind enough to stick a standard tail on them, but my local molecular biology gurus think a little terminal transferase can solve that. But thinking and doing aren’t always the same.

Whale watching -- the looking for extreme reads -- is a sport of some of the most elite labs. Matt Loose's group recently posted a preprint detailing how apparently some huge reads are being broken into multiple reads by MinKNOW -- or something else is going on. The tell-tale are sequential reads from the same pore with very short breaks in between and which align in tandem on a reference genome. Most likely a software issue -- incorrect parsing of the signals -- though the more exotic explanation are pores "playing through" nicks in the DNA. In any case, this brought the whale record to 2.3 megabases (!). But it isn't clear anyone has found a useful whale, one that solves a biological problem. Or even just one that clears an entire large chromosome or a previously unsequenced region such as a centromere. Perhaps LC will bring such news.

Last year ONT presented some interesting details on how little DNA from a library actually makes it to the pore. Clive Brown gave more in one of his webinars. Improving on this would mean success with smaller inputs and potentially higher yields from current inputs. So will any progress be reported?

Polite disagreement between myself and ONT - they think the market is well-served with only 12 barcodes for most kits. I think that’s really low - 96 should be the minimal goal, but more is better so you can freely mix samples (particularly when feeding a PromethION’s maw), plus it makes rolling your barcode sets an option firvdetecting carryover and other contamination.

Basecalling accuracy remains the weak spot of the whole platform. It obviously suffices for many applications, but can be troublesome for de novo assembly. ONT has made steady progress, but perhaps this is plateauing. There is always a risk as well of becoming complacent because the accuracy works in so many situations. Clive's talks on the subject have sometimes veered dangerously close to "we have this solved", but much of the community doesn't agree.

With PromethION bringing the cost of raw bases down to Illumina HiSeq range, planning large de novo or resequencing experiments does get interesting. Illumina gives high accuracy for small features such as SNPs and short indels, but has trouble with long range information such as structural variants. PromethION roughly inverts that relationship. PacBio is more expensive than either but still has the advantage of a well-understood error model (random for simple miscalls, gaussian distribution on homopolymer repeat estimates).

It's a good thing, but with genome sequences such as axlotl and tulip rolling out, the end is probably in sight for setting new records on the size of organismal genomes sequenced and assembled -- we're getting close to the peak of what nature offers.

Will there be any specific information forthcoming on what contaminants cause nanopore sequencing in general -- and the Rapid 1D kit in particular -- to fail.  And just as importantly, which ones really don't matter.  In particular, does RNA contamination really cause problems?  Protein contamination? And what explains samples that fail with Rapid 1D (for example, Mick Watson has mentioned giving up with this kit on sheep rumen microbiome) but succeed with ligation?  It would be unfortunate if people obsess over contaminants that don't really matter  -- or whose impact is indirect.  For example, if RNA is really benign but RNA contamination leads to overestimating DNA concentration and therefore supplying too little DNA into the reaction, that's an issue but not that RNA is a poison.  Once Flongle is online, it should be possible for an interested party -- ideally ONT -- to explore a number of key parameters.  Flongle matters because if these contaminants really kill pores, washing flowcells to search parameter space isn't a desirable tactic.  And it would be good to have actual "worry values" -- akin to LD50s -- for common reagents.  How much Tween or EDTA or any ion matters, rather than an ambiguous statement to avoid high concentrations of X?

Well, let's see what happens tomorrow.

1 comment:

James@cancer said...

WHat si the peak that nature has to offer wrt Genome size?